This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING 1 Automating High-Precision X-Ray and Neutron Imaging Applications With Robotics Joseph A. Hashem, Mitch Pryor, Sheldon Landsberger, James Hunter, and David R. Janecky Abstract— Los Alamos National Laboratory and the University of Texas at Austin recently implemented a robotically controlled nondestructive testing (NDT) system for X-ray and neutron imaging. This system is intended to address the need for accurate measurements for a variety of parts and, be able to track measurement geometry at every imaging location, and is designed for high-throughput applications. This system was deployed in a beam port at a nuclear research reactor and in an operational inspection X-ray bay. The nuclear research reactor system consisted of a precision industrial seven-axis robot, 1.1-MW TRIGA research reactor, and a scintillator-mirror-camera-based imaging system. The X-ray bay system incorporated the same robot, a 225-keV microfocus X-ray source, and a custom flat panel digital detector. The robotic positioning arm is programmable and allows imaging in multiple configurations, including planar, cylindrical, as well as other user defined geometries that provide enhanced engineering evaluation capability. The imaging acquisition device is coupled with the robot for automated image acquisition. The robot can achieve target positional repeatability within 17 µm in the 3-D space. Flexible automation with nondestructive imaging saves costs, reduces dosage, adds imaging techniques, and achieves better quality results in less time. Specifics regarding the robotic system and imaging acquisition and evaluation processes are presented. This paper reviews the comprehensive testing and system evaluation to affirm the feasibility of robotic NDT, presents the system configuration, and reviews results for both X-ray and neutron radiography imaging applications. Note to Practitioners—While looking for ways to improve throughput and increase efficiency in nondestructive imaging applications, the NonDestructive Testing and Evaluation Group at the Los Alamos National Laboratory decided to take a look at automation opportunities. Digital radiography and computed tomography are time-consuming processes, making them ideal candidates for robotic solutions. Radiography applications often require several images to be acquired from different angles and a lot of time they have to be very precise so that the feature of interest is identifiable and the resulting image meets the client’s requirements. With the robot acting as the motion control system, the imaged part can be placed directly in the beam path and oriented in six degrees of freedom. The robot can achieve significantly higher levels of precision than a human and has the Manuscript received August 25, 2016; revised December 20, 2016 and January 31, 2017; accepted February 21, 2017. This paper was recommended for publication by Associate Editor A. Pashkevich and Editor J. Wen upon evaluation of the reviewers’ comments. J. A. Hashem, J. Hunter, and D. R. Janecky are with the Los Alamos National Laboratory, Los Alamos, NM 87545 USA (e-mail: jhashem@lanl.gov; jhunter@lanl.gov; janecky@lanl.gov). M. Pryor and S. Landsberger are with the Nuclear and Radiation Teaching Lab, Department of Mechanical Engineering, University of Texas, Austin, TX 78712 USA (e-mail: mpryor@utexas.edu; s.landsberger@ mail.utexas.edu). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TASE.2017.2675709 ability to adjust the part while the source is active. The system also reduces levels of radiation our staff is exposed to, as the robot is setup to handle radioactive and hazardous parts. Not only does the robot move parts more precisely and with higher resolution than humans, but it also adds additional flexibility in the type and nature of images that the lab can produce. Future work will involve using this system for advanced automated scans such as achieving evenly spaced views around a sphere autonomously, since this system has not yet been used for more advanced scans beyond helical scanning. A tightly linked feedback loop between the robot and imaging code in which the imaging code would autonomously communicate to the robot what additional views are needed to reduce imaging error can also be explored. Index Terms— Autonomous system, calibration, collision avoidance, computed tomography (CT), flexible automation, helical scanning, motion control, nondestructive testing (NDT), path planning, precision movement, radiation damage, radiography, software communication. I. I NTRODUCTION ONDESTRUCTIVE testing (NDT) is a highly multidisciplinary group of techniques used throughout science and industry aimed at evaluation of material properties and detection of defects, both surface and internal, without causing physical damage to the inspected components [1]. Automation of NDT of engineering components represents one of the key objectives of many industries, including the automotive, aerospace, petrochemical, power generation, and nuclear industries. In contrast to manual inspection, it keeps humans away from hazardous material and potentially dangerous work and enables increases in accuracy, precision, and speed of inspection while reducing production time and associated labor costs. The use of robots can provide additional autonomy and flexibility to automated NDT, whereas manual inspection of numerous components or large structures is laborious and time consuming. Automated robot inspection can be beneficial in diverse industrial scenarios ranging from integration of NDT into manufacturing processes of complex geometry components to periodical repair of large structures, such as turbines and aerospace components. Additionally, unusual environmental threats, such as those from underwater oil spills and nuclear power plant accidents, have caused renewed interest in fielding radiography in severe operating conditions. These severe operating conditions pave the way for remote handling systems, where robots are increasingly deployed in remote and hazardous environments, such as those found in the nuclear waste management field and other radioactive environments [2]. The Department of Energy (DOE) has in particular targeted robotic handling of hazardous waste to be an essential element in its efforts of environmental restoration and N 1545-5955 © 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information. This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. 2 waste management [3]. Within the DOE complex, the primary purpose of robots is to replace (or augment) human operators to increase safety without adversely impacting process efficiency. Remote-operated robots allow access and manipulability to areas that would otherwise be inaccessible due to radiation levels, enabling repairs, maintenance work, inspections, or other tasks [4]. A fundamental problem with NDT of manufactured components lies in the process variability. Often parts that are designed identically may have minor to significant deviations from their intended design or even from one part to another. This presents a challenge for precision NDT measurement, since the NDT deployment method must be flexible to accommodate the differences in component shapes. For this reason, NDT inspections are often performed manually by technicians who typically have to position and move a component or inspection probe to achieve proper alignment. This requires trained technicians and results in long inspection times. Achieving a high level of precision and repeatability for a test can be difficult when part alignment is necessary to perform the inspection [5]. Therefore, the fundamental aims of automation within the NDT process are to minimize downtimes due to the higher achievable speed, to minimize variability due to human factors, and to reduce the human exposure to hazardous or physically constrained environments. Semiautomated inspection systems have been developed to overcome some of the shortcoming with manual inspection techniques, using both mobile and fixed robotic platforms. Some NDT techniques, such as video inspection, eddy current testing, and ultrasonic testing, have been readily automated [6]. There are also numerous applications of climbing robots [7] and autonomous mobile robots [8] for inspecting large structures where human access is limited due to space limitations or hazards. Mobile robots carrying inspection tools, such as cameras, 2-D lasers, and IR sensors, have been used for sewer pipe inspection to look for damage or abnormalities [9]. Linear manipulators and bridge designs provide high positioning accuracy [10] to many of these semi-autonomous systems. Typically, these systems are specific machines that are used to inspect identically shaped and sized parts; therefore, they have limited flexibility in the methods they use and what types of parts they can inspect. Automated X-ray test systems for workpieces can be found commercially [11]; however, they come preconfigured for use with specific software and hardware. Despite these previous efforts, challenges remain to be addressed before fully automated, high-precision NDT inspection becomes commonplace. The key challenges include flexible trajectory planning, integrated NDT data collection, and achieving high repeatability and precision measurements. The result of NDT imaging inspection requires a flexible and extensible approach that has the flexibility to allow future changes in the path planning to accommodate different parts and techniques. This effort focuses on flexible motion control system for NDT applications in radioactive and other hazardous environments that are complex, requiring multidisciplinary knowledge in order for the motion system to complete its required task(s). Flexible NDT is necessary to address the vast majority of NDT IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING applications which are experimental, low (or often singular) batch, and/or require in situ modification of test parameters. As computer vision and automated inspection techniques improve, automated imaging inspection and interpretation can become a reality. Research and development is, however, ongoing into new approaches that can be used to aid X-ray interpretation [12]. Today, any particle type can be combined with an increasingly wide range of digital detectors to image almost any conceivable object in extreme environments. As the industrial radiography sensitivity and resolution requirements increase, along with the range and complexity of items to be inspected, high-precision positioning systems become necessary. Imaging techniques rely heavily on positioning the part in a precise location inside the X-ray beam and inline with the imaging detector, not just once, but many times (up to thousands of related but distinct positions). Typical imaging techniques include: static radiography, computed tomography (CT), and helical scanning [13]. The NDT and Evaluation Group at the Los Alamos National Laboratory (LANL) has implemented various linear and rotary motion stages to position parts for digital radiography (DR) and CT [14]. Additional techniques such as helical scanning and in-motion radiography have been implemented [15]. CT scan requirements are a good example of the high precision and repeatability requirements for X-ray imaging. CT typically requires one view per pixel of maximum region of interest width. In tomography, a variety of artifacts may be present in projection sets that propagate errors back into the reconstructed image. If fewer projections of the object are captured, the image will have more reconstruction artifacts and poorer resolution, boundary definition, and uniform voxel (i.e., the 3-D analog of the pixel in a 2-D image) spacing. Thus, it is necessary to have uniformly spaced view angles. With CT scans, it is important that images be precisely located relative to each other. A six-axis positioning system, along with a digital detector would yield the following benefits. 1) Enhanced Worker Safety: Less frequent access to the beam area reduces the probability of an accidental worker exposure, and helps maintain as low as reasonably achievable (ALARA) dose by keeping distance between parts and workers [40]. 2) Positioning Flexibility: A robotic positioning system allows the radiographers to get images from any angle, and parts can be repositioned remotely to better capture the desired image. 3) Real Time Remote Positioning: Coupled to a real-time digital imaging system, radiographers can remotely position the item to better investigate newly discovered areas of interest. In the sections to follow, we describe the robotic system, how it is used, the integration between the components of the entire imaging system, and evaluate the resulting X-ray and neutron radiographs. Robotics allow for high-mechanical rigidity and repeatability, configurable imaging motion planning, six DoF programmable positioning of payloads with both course and fine movement capability, and kinematic model visualization. For example, the robotic arm used in this paper (see Section II-C This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. HASHEM et al.: AUTOMATING HIGH-PRECISION X-RAY AND NEUTRON IMAGING APPLICATIONS WITH ROBOTICS 3 Fig. 1. Part positioning system installed in the open-air X-ray bay at LANL. The setup consists of a computer controlled robot and a flat panel digital detector that communicate with one another autonomously. for more specifics on the robot model) is capable of moving a 5-kg payload over a near-arbitrary path within a volume of roughly 0.4-m radius, while achieving a repeatability of 17 µm. Other NDT applications such as eddy current [16] or ultrasonic inspection of aerospace components [17] have implemented the use of a robotic arm for sample exchange and/or part positioning. However, these applications do not necessarily require the same level of precision as the imaging applications required in this paper, with an example shown in Fig. 1. Fig. 2. Sample handling process. The first “Wait for input” block allows the user to either choose a sample to pick up or turn the system OFF, while the second “Wait for input” block asks the operator to enter the desired rotation and offset amounts. II. M ETHODS AND G ENERAL D ESIGN C ONCEPT The application task is to autonomously image different parts, one-at-a-time, and with high precision. The application begins with the objects either randomly distributed or in a predetermined position in the robot’s workspace. If the objects are randomly distributed, an imaging system is used to determine their location using sensor information from a depth camera [18]. A schematic of the generalized cycle of sample handling is shown in Fig. 2. The process shown here represents the X-ray system. The X-ray generating device (XGD) and digital imaging detector are also included in Fig. 2. The robotic system deployed in the nuclear research reactor follows a similar sample handling process, except that the XGD is replaced by the neutron beam and the detector consists of the scintillator-mirror, camera system. The neutron imaging system is described in detail in Section III-B. For both X-ray and neutron systems, the task is always performed in simulation prior to hardware execution to make sure the robot completes the task as expected and safely. A risk analysis was carried out, identifying breakdown recovery as a particular area of concern because of the potential handling of radioactive material and the restrictions on personnel access as radiation levels increase. The robotic system therefore includes features to either avoid foreseeable problems or to enable recovery in the event of failure when the robots are handling highly radioactive parts. One example of problem avoidance is the implementation of collision avoidance. Another example is robot joint current monitoring during operations; if a joint current exceeds its threshold value, the robot will stop. This monitoring allows detection of problems before the robot generates large forces on fixed surfaces or other equipment in the area. This ensures recovery is not hampered by the robot having to “cut out” after exceeding the maximum joint current limit. The system permits the operator to easily cancel autonomous execution at any time and revert to teleoperation, or issue new high-level motion commands, such as “stow robot behind shielding” or “place part down,” without the need to restart any hardware or software. The user is kept in-theloop, so one is able to intervene in the case of an emergency. This type of flexibility permits operation under transitional autonomy, and is not possible under the traditional teach/do paradigm of industrial robot programming. A. NDT Imaging Requirements The goal of performing flexible imaging and maintaining required positioning accuracy require that positioning is correctable in translation (x, y, z) and in pointing angle (Rx, Ry, Rz). Additionally, multiple axes need to be aligned (Fig. 3) and the imaged part’s axis of rotation needs to be aligned with the system’s z-axis (i.e., the detector) and the robot axes. According to LANL radiographers, the most precise of product specifications call out 0.05-mm minimum feature requirements to be measured for radiography inspection in the LANL Plutonium Facility. These features are measured to 0.025 mm. Vibrations cannot exceed 6 µm (i.e., one quarter of the maximum requirement) while the part is held in order to allow accurate measurements. Also, the tilt of the part relative to the X-ray beam needs to be known to within 0.1°. These requirements represent typical requirements for NDT applications surveyed at LANL. This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. 4 IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING Fig. 4. 3-D blank beam image showing beam structure due to neutron guides. Fig. 3. Robot, X-ray source, flat panel detector, and coordinate system definitions. Imaging sphere is not drawn to scale. B. Summary of Utilized Systems The key hardware elements utilized to be developed in the automation imaging system include the following: 1) seven axis industrial manipulator and gripper capable for part manipulation; 2) X-ray source; 3) neutron source; 4) operational software for flexible automation capability and peripheral integration including vision. helps to maintain positioning accuracy and minimize errors throughout the entire radiograph exposure time. This level of correction is not possible with conventional stacked-axes positioning systems that typically have fewer than six DoF as well as a small reachable workspace. Additionally, stacked motion stages can only move a single axis at a time. A robot can be taught and/or commanded to perform a sequence of complicated moves. In both robotic and motion stage-based systems, the volume occupied by the item inspected and that items swept volume path must also be accounted for. An analysis of the robot’s workspace and collision avoidance is described in Section II-F. D. X-Ray Source and Digital Detector C. Seven-Axes Robot for Part Manipulation To address the mechanical challenges posed by the highprecision requirements in NDT, we have implemented the use of a seven-axis serial robot with a 5-kg payload (model Yaskawa SIA5 and FS100 controller [19]). A Robotiq gripper (both three-fingered [20] and two-fingered models [21] were used). The tip of the end effector (EEF) is also commonly referred to as the tool center point (TCP). The robot controller uses the Denavit–Hartenberg parameters [22] to define the kinematics of the robot, which allows the part to be spatially oriented. The choice of this industrial-grade, robotic arm allows for the reduction of time needed for inspection, high precision, affordability, reliability, as well as integration with open source software which simplified the integration of peripheral devices, vision, and other supporting components. The coordinated motion and part offsets must provide the ability to maintain positional accuracy for various parts, accounting for varying part sizes and weights. The robotic software combined with the well-developed robotic controller has integrated timing mechanisms that provide input and output signals to easily coordinate the X-ray source, flat panel detector, and other measurement equipment precisely at desired times of the imaging application. Polar rotation of the part is accomplished by inputting the part offset distance and tilt relative to the robot’s tooling offset and then selecting a desired part rotation. Indents in the robot’s gripper provide for repeatable part positioning. The coordinated motion of the robot allows for full spatial error correction, including position and orientation. This The system was installed in an open-air X-ray bay at LANL with a 225 keV microfocus X-ray source. A custom Varian 2520 (25 cm × 20 cm) amorphous silicon flat panel digital detector [23] that is tolerant to high-energy X-rays was used to acquire digital radiographs. All radiographs throughout this work were imaged with the XGD set at 135 keV and 66 mA. The exposure time was set to 0.5 s for each image (i.e., 2 frames/s) with 50 frames averaged for each image. For CT and helical scans, the number of averaged frames was reduced to 10 to decrease scan time. The focal spot size was set to 7 µm and a 6:1 magnification was used (source to object distance was 280 mm and source to detector distance was 1701 mm). E. Neutron Source and Scintillator-Mirror-Camera System Neutron imaging was performed at the University of Texas (UT) Austin’s TRIGA Mark II research reactor. In the beam port, there is a thermal neutron flux of 5.3 × 106 n cm−2 s−1 and thermal-to-epithermal ratio of 8.1 × 104 ± 10% n cm−2 s−1 at a reactor power of 950 kW. Neutrons are directed from the reactor core to the beam port through a neutron guide, which reflects neutrons in a manner that is analogous to optical reflection. The neutron guide acts as a neutron filter; only neutrons of certain energy are efficiently transported down the guide. Neutron guides work best for lower energy neutrons as the reflectivity of higher energy neutrons is low. From Fig. 4, one can see that there is structure in the beam, which is due to the neutron guides. When calibrating images this structure is accounted for by This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. HASHEM et al.: AUTOMATING HIGH-PRECISION X-RAY AND NEUTRON IMAGING APPLICATIONS WITH ROBOTICS Fig. 5. 5 Neutron imaging acquisition setup. normalizing the blank beam measurement. The beam is a very clean neutron source and is also a divergent beam. A scintillator-mirror-camera system (Fig. 5) was utilized to acquire digital radiographs. This allows the operator to know what the radiograph looks like in real-time and allows for adjustments in part positioning to be made online. The neutron imaging detector’s pixel pitch, which is the physical distance between the pixels in the imaging device, is 35 µm. The charge-coupled device camera had a 1004 × 1002 active pixel area on the chip, with a pixel size of 8 × 8 µm, and a lens magnification of 1:1.14. The scintillator used was a copper, aluminum, and gold doped 6 LiF ZnS neutron detection screen. The overall detector resolution of the system is limited by the scintillator resolution (i.e., 6 lp/mm), which is equivalent to the minimum resolution of 80 µm. For future work, the modular transfer function should be calculated for the neutron imaging system; however, the focus of this work is to show the functionality and feasibility of the incorporation of robotics into the imaging system. The estimated thermal neutron capture efficiency is 42% at a scintillator thickness of 0.45 mm, a 6 LiF:ZnS mix ratio by weight, and a 6 Li atomic volume density of 12.9 atoms/ccx1021. [24] The reaction that occurs is 6 Li + n → 4 H e + 3 H + 4.8 MeV, where the ejected triton interacts with phosphor in the scintillator to create a scintillation event. A stainless steel enclosure, along with lead bricks and lead blankets, surrounded the enclosure to shield against X-ray hits and background noise. Note that CT scans were not performed using this neutron imaging system, since exposure times to obtain a single image ranged from five to ten minutes. F. Robot Software The robot operating system (ROS) was used for operating the robot, integrate all peripheral devices, and perform many of the necessary calculations, including object recognition, pose estimation, trajectory generation, and collision detection [25], [26]. The reachable workspace of the SIA5 was calculated and is shown in Fig. 6 with a collision object added to the robot’s workspace. This demonstrates the collision avoidance Fig. 6. Collision object, which is the “gripper holder” is shown in purple in the robot’s workspace (left). Isometric view of the robot’s workspace (center). Top–down view of the workspace with the location of collision object shown in the dashed red circle (bottom). Note that the lines in 2-D are projected from above in the top–down view. capability of the software and shows the workspace boundaries of the manipulator with a Robotiq two-finger gripper attached. From Fig. 6, it is evident that the range of motion from the axis of rotation around the z-axis to the last point of the EEF, in the y-axis direction is roughly 0.72 m. It should be noted that this range is only theoretical. To acquire the SIA5’s workspace, MATLAB’s Robotic System Toolbox [43] was utilized to record the EEF path at 10 Hz as the robot moved to random locations. The EEF position was recorded by obtaining the transformation between the robot’s base and the robot’s palm. A similar method can be used to obtain the orientation of the EEF at each location, if desired. Joint interpolated motions were used to complete each move. It is apparent that the robot automatically avoids the collision object, as shown by the lack of motion inside the dashed red circle in Fig. 6. There were a total of 400 000 moves tested and no failures were identified—less than 1 in 400 000 failures (less than 2.5 per million moves) for the point at the EEF. For actual experimental testing, the move can be simulated first and then stored for a particular motion plan needed for imaging applications. G. Achieving Small Angular Motions for Tasks Requiring High Precision CT and helical scans [38] can require more than 4000 images, depending on the part geometry and dimensions, taken at sequential angular steps. This requires the robot to execute joint angle movements of less than 0.1°. In order to do this, the software computes a very short trajectory for each step with four points along each trajectory; the first and last are the starting and ending point respectively and the middle two are determined by linear interpolation. Without performing this trajectory, the minimum increment a robot can move via a This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. 6 IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING simple joint motion command (i.e., arm->move()) is 0.287, where arm is the name of the MoveGroup class that will be controlled and planned for, and move() is a function within MoveIt that moves the robot to a specified joint position [42].1 A 65.2% decrease (from 0.287° to 0.1°) in the minimum joint angle step value was achieved by using this trajectory algorithm. The same mini trajectory used to achieve small joint motions can be used to give micrometer level commands to the robot for Cartesian motions in ROS. H. Part Alignment Alignment of item center or other point of reference can be one of the most challenging and time consuming aspects of NDT imaging. This can be especially true for high-resolution imaging, where digital panels and magnification can lower the required alignment tolerances to hundreds of micrometers. Having to keep track and maintain positions of the robot, gripper, and part to these tolerances is a challenging requirement. To determine and maintain mechanical relationships among these components offset frames between the robot’s EEF to the gripper and to the desired part pose must be calculated. To accomplish this, the robot kinematics must be calibrated so that the robot knows where it is and where the held part is relative to the measurement coordinate system. This entails teaching the robot where the part is in relationship to its native kinematic model, which allows the robot to be commanded in the part’s coordinate system. To do this, transformation matrices are derived, which allow the offset frames to be directly interpreted as a set of movements in the robot’s native coordinate system. The part frame is the robot’s EEF frame plus a translational and rotational offset. The position of the part frame, PF (x, y, z), relative to the tool-center-point frame (located at the robot’s wrist faceplate), TCP F (a, b, c), is mathematically determined by the 4 × 4 homogeneous transformation matrix represented by   ax bx cx px  ay by cy py  Part   (1) TCP R =  a b z cz p z  z 0 0 0 1 where the coordinates of vector p = ( px , p y , pz ) represent the location of PF and the coordinates of three unit directional vectors a, b, and c represent the orientation of TCP F . The inverse of the transformation matrix in (1) represents the position of frame TCP F to frame PF . The orientation coordinates of frame TCP F in (1) can be determined by   ax bx cx  a y b y c y  = Rot(z, θz )Rot(y, θ y )Rot(x, θx ) (2) a z b z cz where transformations Rot(x, θx ), Rot(y, θ y ), and Rot(z, θz ) are pure rotations of frame TCP F about the x-, y-, and z-axes of frame PF with the angles of θx (yaw), θ y (pitch), and θz (roll), respectively. 1 This minimum joint value increment was determined experimentally using an SIA5 manipulator, FS100 controller, ROS Hydro, and Ubuntu 12.04. More work needs to be done to determine if this is an ROS, FS100, SIA5, or MoveIt [42] limitation and why. Fig. 7. X-ray system setup with integration of robot, ROS, gripper, controller, and image acquisition device shown. For the neutron system, the flat panel is replaced with the scintillator-mirror-camera system described in Section II-E. After determining the EEF-to-part offset frame, the relationship between the robot and different coordinate systems can be determined such that the part remains in the beam path along the z-axis (Fig. 3). This alignment allows the robot to rotate the part while maintaining the part’s alignment in the beam and field-of-view of the detector. Knowledge of the part volume and volume sweep is important and can be accounted for by modeling the part in ROS a priori. I. Coordination Between Robot, Imaging Detector, and XGD or Neutron Beam In order to accomplish motion planning, robot trajectory execution, and integration with other hardware, it is necessary to have a suitable framework. The developed software must simplify robot integration with other NDT hardware components (the imaging detector and XGD or neutron beam). The software developed in this paper also allows for the integration of image acquisition devices to make the motion and image acquisition autonomous. Whereas the hardware system is composed of off-the-shelf components, the software system is custom-built to provide the flexibility required to automate integrated tasks. Separately, the tasks themselves are not difficult, but integration into a single system is nontrivial. Fig. 7 shows the major software and hardware system components and interactions. ROS includes the robot software and application code (e.g., pick, place, and so on). The software (hardware drivers, algorithms, etc.) is organized as a set of nodes that communicate via a standard messaging protocol. This custom LANL automated radiography system has been developed to simplify application-level programming. System architecture was developed for integration of radiography imaging with a robotic manipulator where images are automatically acquired at each step in the scan, through communication between the robot’s motion control and image acquisition. A modular architecture facilitates different imaging devices integration and utilizes a Python script to communicate between the robot and flat-panel detector to close This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. HASHEM et al.: AUTOMATING HIGH-PRECISION X-RAY AND NEUTRON IMAGING APPLICATIONS WITH ROBOTICS the loop between the robotic motion and imaging acquisition device. ROS allows for communication between the robot and imaging devices using scripting languages (C++, Python, and MATLAB) on the same PC. The code to run the robot is written in C++, and the flat panel acquisition software is written in Python. ROS has tools to effectively communicate between the multiple processes both simultaneously and sequentially. ROS creates a publisher “talker” node that broadcasts a message to either the flat panel detector when the system is ready to acquire a radiograph. A master node tells the flat panel detector node that is subscribed to an acquire_image topic when to acquire an image. This topic tells the master node whether to continue with the robot application if the image acquisition has finished, or to wait if the image acquisition device has not yet completed the exposure time or completed the saving process of the radiograph. J. Measuring Radiation Damage Radiation exposure can result in damage to the robot and other components. Radiation can cause energy deposition in materials that can cause mechanical and electrical changes, such as weakening to material structure due to vacancies and irregularities and reducing semiconductor effectiveness due to formation of ion pairs from deposited energy. Monte Carlo tools like MCNP [27] can enable us to easily perform the high-fidelity calculations necessary to determine the neutron damage rate. High enough levels of neutrons or photons will eventually affect the reliability of electronic components. Thus, radiation tolerance is critical to the reliability of the imaging process. The sensitive components installed on advanced manipulators can be divided into three categories: 1) the drives (usually electrical actuators with bearings, gear boxes and position feedback devices); 2) the sensors (distance and force sensors, and cameras); and 3) the cables and other communication devices (including line drivers, multiplexing circuits, analog to digital converters, radio links and even the preamplifiers needed for some sensors). For each category, the radiation hardening level required will depend on their location with respect to the radiation sources (near the EEF or near gantry tracks or walls) and on their frequency of use (e.g., a tool used a small number of times, compared with protection systems in use permanently) [28]. Papers detailing this work and the results can be found in [29] and [30]. In summary, Displacement Per Atom (DPA) rates in the SIA5 robot were calculated using MCNP. For this work, the neutron spectrum was simulated and based on the same beam port that the robotic system was deployed in at UT Austin’s TRIGA Mark II research reactor. This type of DPA rate calculation can be applied to determine the radiation damage to robots and other objects in other radioactive environments and applications. For comparison, the DPA rates determined in this work were similar to those found in thermal reactor materials [31], which is expected since the TRIGA reactor is one. III. R ESULTS The previous section (methods and general design concept) provided detail on the methods used to determine the 7 reliability, safety, optimization, communication, and execution of a robotic system for NDT imaging purposes. This section delves into the details of the system implementation and uses the methods presented in the previous section to perform several experiments that test and validate the advantages of using flexible automation for nondestructive imaging purposes. The implementation of the system is described, showing how the methods presented are used in conjunction with other hardware and software tools to provide a functional autonomous radiography and/or CT system. Two main application areas are discussed that show the robot performing neutron imaging at a TRIGA Mark II research reactor at UT Austin and X-ray imaging at a high energy X-ray source at LANL. This application area demonstrates how the work presented in this document supports the feasibility and necessity of a robotic system for nondestructive imaging by exploiting the flexibility of robotics to gain efficiency and adhere to ALARA principles. All of the grasp, motion planning, and image acquisition communication in the following applications are performed in ROS, demonstrating how ROS simplifies integration of complex robotic systems under a common operating framework and how ROS allows incorporation of recent research advances into deployed systems. A. Metrology In depth repeatability and vibration tests have been performed on the robotic system used in this paper [30], [32] and demonstrate that the robot’s repeatability and vibration mitigation are sufficient for the required imaging tasks. For the robot’s mechanical parameters assessed, the tests performed showed that the repeatability is 17 µm with a slight dependence on speed. Faster robot movement speeds (100% instead of 25% of v max , which is approximately 2500 mm/s for linear moves [19]) result in slightly worse (less than 25 µm) repeatability results. Having high repeatability allows the robot to position the part in the beam in the same desired pose each time. Yaskawa [19] states a repeatability value of 60 µm. However, asserted capabilities from robotic companies tend to be extremely safe, since Yaskawa’s declared repeatability value is taken at maximum payload and maximum speed [33]. For actual hardware implementation, the robot’s velocity is limited to 20% of its maximum velocity and acceleration limit. This limitation is defined by the implementation in this paper. Two types of vibrations that are explored in Hashem et al. [30], [32] are static and tracking vibration, with static vibration found to be within the 6-µm requirement. Static vibration is the amplitude of impact of vibration on the EEF position while the robot is not moving. Tracking vibration is the amplitude of impact of vibration on the EEF position while the robot is moving. Tracking, which is the ability to follow the exact same EEF path, is also explored. The repeatability of the robot can also be determined using actual radiograph images. This test was performed using a 225-keV microfocus with a Varian amorphous silicon flat-panel detector (PaxScan 2520, Varian Imaging Products, Palo Alto, CA). The Varian panel’s pixel pitch is 127 µm at 1:1 magnification. In ordinary X-ray radiography, the resolution in the captured images is 25 µm for film and 100 µm– This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. 8 1 mm for digital. The microfocus allows imaging down to the 2–3 µm level. Therefore, it is important for the robot’s repeatability and resolution to be on this level as well. The distance between the source and object should be minimized to increase magnification so that micrometer level motions can be resolved. However, decreasing the source to object distance increases focal spot blur. Lower energies can help reduce image blurring while longer exposure times improve statistics. Blurring in the captured images should be minimized so one can distinguish the micrometer level differences between images. Ten radiographs were taken of a spherical ball bearing (BB) attached to a plastic screw held by the robot. The robot completed a repeatable motion (performing both joint and Cartesian motions), with each image taken at the completion of the repeated move. Images were compared in terms of pixel intensity and location relative to other images. The differences in pixel intensity relate to the differences in the location of the BB. It was found that there was approximately a 1.3 pixel difference in BB location along the x- and yaxes between the images. For this test, the source to object distance was 171.45 mm and the source to detector distance was 1701 mm, so the magnification or zoom factor was 9.92. Therefore, the effective system pixel pitch was 12.8 µm (i.e., 127 µm pixel pitch divided by 9.92). The corresponding repeatability value would then be ±17 µm (i.e., 12.8 µm times 1.3 pixel value difference). This compares well to the ±17.9 µm repeatability value obtained using the dial indicator with the robot moving at 25% maximum speed. It is important to use the void setGoalTolerance()function in ROS to set the error tolerance to a sufficiently low value. For joint state goals, this will be the distance for each joint, in the configuration space (radians). For pose goals this will be the radius of a sphere (m) where the EEF must reach. Resolution for a robotic system is the minimal commanded step for a joint. The resolution value is the smallest incremental move that the robot can physically produce. To test the resolution of the robot’s EEF, the robot was commanded to move the minimal step in Cartesian space by computing a mini trajectory in ROS. The same microfocus system was used to acquire the images as described earlier. An example test is shown in Fig. 8. For the resolution test conducted, the distance from the source to the object was 12.7 cm and the distance from the source to the detector was 167.6 cm. Therefore, the magnification was 13.2 and the effective system pixel pitch is 9.62 µm (i.e., 127 µm divided by 13.2). The differences between the images are shown in Fig. 9. The x-direction is facing up in the images. There was approximately a 3.5 pixel difference between the two images. This value needs to be divided in half to get the actual distance traveled because the differences in both images are highlighted. Therefore, the actual distance traveled was 1.75 pixel, which relates to a 16.8 µm resolution (i.e., 9.62 µm times 1.75 pixels). The joint resolution of the robot will be lower than this value since the resolution possible for the EEF is a function of the joint resolution and the configuration of the robot. IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING Fig. 8. Resolution test. The robot was commanded to move a minimal amount in Cartesian space along in the positive X-directions while holding an object with two BBs attached. Initial position (left) and final position (right). A scale bar is shown in the top-left corner. Fig. 9. Differences between BB locations on two resolution test images. The dashed rectangle (not to scale) (left) shows the zoomed-in section with the difference in pixels measured (right). Scale bars are shown for both images. B. Neutron Imaging Measurements To illustrate the advantages of using a robotic manipulator with neutron imaging, mock-up depleted uranium fuel rods, each consisting of five pellets prepared from urania (UO2 ) powder, were characterized by thermal neutron radiography. To simulate cracks and voids resulting from irradiation and burn-up in a fuel pin, tungsten and gadolinium inclusions were embedded in the mock-up pellets. These rodlets contained defects similar to those seen in irradiated fuel rodlets. They can be used to establish sensitivity for density, visualization of voids/cracks, and inclusions of different materials. The SIA5 manipulator handled the fuel rods and provided advanced and flexible motion capabilities that would be difficult to achieve with linear and rotary motion stages. The goal is to characterize irradiated fuel pellets, as well as to offer better guidance for more expensive destructive examination. By imaging fuel rods, one can see the effect of the development of irradiation and burn-up damage in nuclear fuel over time. Since the technique is nondestructive, the time evolution of fuel rod damage can easily be measured this way. With the ability to predict how the composition and structural integrity of fuel pellets evolve during their duration in a reactor, one can improve the performance of nuclear modeling codes such as MARMOT [34]. This will accelerate the understanding of processes occurring during irradiation and ultimately improve nuclear fuel. This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. HASHEM et al.: AUTOMATING HIGH-PRECISION X-RAY AND NEUTRON IMAGING APPLICATIONS WITH ROBOTICS Fig. 10. Neutron transmission images of three mock-up urania fuel rodlets with engineered flaws, including gadolinium and tungsten inclusions. The black ovals indicate gadolinium inclusions, and the red oval indicates a tungsten inclusion. A steel container encloses each of the fuel rods, and a spring is seen in the top of each assembly. A scale bar is shown in the top-left corner. Neutron imaging consisted of a radiograph of three fuel rods with the primary focus on the fuel pellet region. Exposure times ranged between five and ten minutes for a single image. Multiple experiments were conducted including the following. 1) Rotating each fuel rod to various orientations. 2) A vibration analysis (i.e., comparison of the resulting radiograph with the robot holding and not holding a fuel rod). 3) A repeatability test. This test consisted of taking an image with the robot holding the fuel rod in a specified location in the beam, moving the robot away from the beam, bringing the robot back to the same specified location, taking another image, and then comparing the two images using MATLAB’s Image Processing Toolbox [35]. From these comparisons, it was shown that the repeatability in the fuel rod’s final location was ∼0.025 mm. More information on this test can be found in Hashem [30]. 4) A CT scan of one of the fuel rods. Neutron CT is a process by which the 3-D neutron attenuation values throughout the object are obtained. This process requires taking 2-D neutron radiographs of the object while it is rotated around 360°, with images taken at certain points within the scan. Recon [36], a CPU-based reconstruction algorithm that uses a standard Feldkamp filtered back projection method, was used to reconstruct the 3-D map. 5) A helical scan, which is explained in Section III-C. 6) Radiographs of the fuel rods at various orientations. Example results of the neutron radiographs of the three fuel rods are shown in Fig. 10. Darker regions represent materials/ areas with higher neutron attenuation coefficients than lighter regions. Gadolinium and tungsten inclusions appear darker than the depleted uranium (d-UO2 powder). The gray regions indicated uniform d-UO2 . Thermal neutron tomography identifies flaws in the composite pellets whereas areal density fits 9 Fig. 11. Subset of helical scan of a Maglite completed using the robotic system coupled with the image acquisition system. The robot’s gripper can be seen in the first three radiographs, however, it does not interfere with the resulting radiographs since we are only interested in the bulb (top) portion of the part. to the 238 U demonstrate density uniformity. The gaps between rodlets are visible in the radiographic images. Voids or chips on the outside of the pellets are visible (appear black). These radiographs clearly show the capability of the robot to perform neutron radiography tasks. The integrated system has successfully demonstrated imaging of the mock-up uranium fuel rods with the necessary precision and repeatability. These demonstrations serve as a proof-of-concept that flexible automation and robotic technologies developed in research laboratories can be valuable for advanced nondestructive imaging abilities and applications. C. X-Ray Imaging Measurements A seven-axis robot arm can be configured for any number of imaging scan types, including CT, helical scans, and in-motion radiography [13]. Thus the robotic system is an allin-one system to perform different X-ray imaging scan types. An example of this is shown in Fig. 11, where X-ray images were taken of a helical scan of a Maglite [37] performed with the robotic system. In order to perform the helical scan, the robot simultaneously translated the part vertically and rotated the part at each step in the scan. Helical scans are implemented to decrease scan time when the imaged object is longer than the detector’s field-of-view. It is also useful when using a point source, where the source strength is not as strong near the top and bottom of the image. The capabilities and application process for helical scans are well documented by Silverman et al. [38]. The six images below in Fig. 11 (left to right) show the progression of a helical scan of the Maglite. These images are just a subset of the entire helical scan from the beginning to end of the scan (rotated from 0° to 360° and translated 7 cm in total). Each progressive image in the radiographs below show the part rotated 51° and translated 1.17 cm. The two-fingered Robotiq gripper is visible in the first five images but does not obtrude the critical features of the Maglite. This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. 10 IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING Fig. 12. Reconstructed results of a CT scan of a Maglite obtained with using the robot and rotary stage to rotate the part. A cross section slice of each method is shown below the respective radiograph. A scale bar is shown in the top-left corner. Fig. 13. Heat-source radiographed in normal configuration (left) and radiographed in “pole shot” configuration (right). A CT scan of the Maglite was also performed using the robotic system as well as a standard rotary stage [39] for comparison. Fig. 12 shows the reconstructed CT images and cross-sectional slices of the part that were acquired using both systems. The scan performed using the robotic system consisted of a 360° scan with 600 projections. A similar scan was acquired using the rotary stage to rotate the part; however, 1000 total projections were acquired. The same number of projections could have been acquired using the robotic system, but was unnecessary to see that the results are comparable. A mock heat source, approximately 1 in in diameter, was also radiographed using the 225 keV microfocus X-ray source to demonstrate the system’s high precision capability. There are many instances in radiography where specific views of a part are needed. One common view is a pole-shot of an object, where the beam goes through the part from the top to bottom. For example, a pole-shot radiograph of the heat source is seen in Fig. 13. This radiograph was acquired using the robotic system as shown in Fig. 14. With this system, one can acquire various views and angles of a part, and even place the part back down and pick it up in a different orientation if a specific view such as a pole-shot is required. Without the robotic system, the operator would have to shut down the radiation source and manually reposition the part. For some specific shots, such as pole-shots, a fixture may even need to Fig. 14. Pole-shot demonstration. (a) Robot picking up the heat source from the top and (b) placing it in the X-ray beam in a regular orientation. (c) Robot then places the part back down, (d) approaches new pickup position, and (e) picks it up from the side. (f) Heat source can then be positioned in the beam to get a pole-shot radiograph. Fig. 15. Precision alignment capability of the robot when a part’s weld needs to be aligned with the detector. The radiograph on the left shows the weld not aligned and the radiograph on the right shows the weld aligned with the detector. The intensity transmission plots are shown to the left of their respective radiograph. be machined to hold a part in place so that it does not move while imaged. This work focuses on asking industrial robots to do flexible tasks on a small scale with the precision required for nontrivial NDT tasks in a DOE complex. A demonstration of this capability is shown in Fig. 15, which shows the X-ray radiographs and intensity transmissions of the heat source. The weld in the heat source must be perfectly aligned with the detector This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. HASHEM et al.: AUTOMATING HIGH-PRECISION X-RAY AND NEUTRON IMAGING APPLICATIONS WITH ROBOTICS that, when performed manually, is a time-consuming and frustrating exercise in trial-and-error. The intensity transmission histograms are used to align the part autonomously. On the left of Fig. 15, the two X-ray intensity transmission peaks marked are produced as X-rays penetrate the weld area of the heat source: one for the front side of the weld and one for the back. The peaks result as the X-rays are attenuated to a lesser extent where the heat source is welded together as the weld appears in the radiograph as a material of slightly different density. To properly align the heat source so that the imaging system only sees a single peak of X-ray intensity transmission, the robotic system must align the part as shown on the right of Fig. 15. To accomplish this task, a feedback loop is made between the robotic system and the imaging system that continuously takes digital radiographs in real-time. The robotic system, adjusts the part until a single X-ray intensity transmission is observed in the region of interest. When the weld is perfectly aligned, the two peaks will converge into a single peak. A real-time positioning system with digital feedback allows this alignment to be achieved quickly without manual repositioning of the part. IV. C ONCLUSION We have demonstrated the utility of using a robot for positioning of radiography samples to reduce worker exposure and achieve real-time, arbitrary six-axis sample positioning. The performance gains achieved by the robot enable advanced radiography techniques such as helical scanning, CT, and alignment of welded components. These demonstrated using an industrial robot deployed in both a nuclear reactor beam port and a high-energy X-ray bay. This effort is a collaboration between AET-6 at LANL and UT Austin. The use of robotics and automation reduces personnel dose, improves precision alignment, and improves throughput. Automation has the additional potential benefit of improving part throughput by obviating the need for human personnel to move or exchange parts to be imaged and allowing for flexible orientation of the imaged object with respect to the X-ray or neutron beam. We have presented a new approach for X-ray imaging using robotics that opens the door for new possibilities for NDT imaging. The ability to have six DoF control while keeping alignment repeatability tolerances to less than 20 µm means that configurable imaging scans are not only definable but maintainable. The configurability allows for an automated imaging system that has the same functionality as multiple separate imaging systems and one that can accommodate a large variation of part form factors. Furthermore, the linear and rotation positioning capabilities needed for magnification changes and part tilt can also be achieved with this approach. The use of commercially available robotics allows for scaling of this system relatively easily as there is consistency in robot controllers for a large variety of robot models that support ROS [25]. We have shown that the stable positioning for X-ray imaging can be performed to at least the required 25-µm absolute repeatability level defined by LANL engineers. This positional error may be further reduced by use of a laser tracker that monitors and records positions for fine robot position correction as shown by Gordon et al. [41]. 11 The system can run autonomously as well as in teleoperated mode, and it gives technicians the ability to position samples in real time, in situ, and arbitrarily in 6-axes without entry into potentially high radiation areas. The ability to arbitrarily position a sample in space ensures that the radiographer can obtain exactly the image they need without the need to design and fabricate new fixtures or tools. The estimated time saved over manual operations is at least 3 min/view. For example, if 1000 parts are imaged, 50+ h of labor is eliminated. This does account for the time it takes for reentry into radiation areas and inevitable small motion adjustments. The robotic system also extends operation and the use of multiple shifts. Future work will focus on improving the feedback loop between the robot and CT processing algorithms, which allows for additional imaging techniques, applications, improved radiograph quality, and reduced throughput time. For example, this system can achieve evenly spaced views around a sphere autonomously. Also, after an initial radiograph is acquired, the CT processing algorithm (or operator) could request additional views to reduce error and achieve more optimal radiographs. Sample holder trays to hold and organize parts will be printed via additive manufacturing or fixture free grasping systems under development in a parallel effort can further reduce integration efforts. Gripper interlocks and grasp validation techniques can prevent the release of a part except in designated process locations for additional safety measures if needed. R EFERENCES [1] L. Cartz, “Nondestructive testing,” ASM International, Materials Park, OH, USA, Tech. Rep., 1995. [2] G. S. Sundar et al., “Design and developments of inspection robots in nuclear environment: A review,” Int. J. Mech. Eng. Robot. Res., vol. 1, no. 3, pp. 400–409, 2012. [3] U.S. Dept. Energy, “Environmental restoration and waste management robotics technology development program, robotics 5-year plan,” DOE, Washington, DC, USA, Tech. Rep. DOE/CE-0007T, 1990, vol. 3. [4] K. Nagatani et al., “Emergency response to the nuclear accident at the Fukushima Daiichi Nuclear Power Plants using mobile rescue robots,” J. Field Robot., vol. 30, no. 1, pp. 44–63. 2013. [5] T. Sattar, “Robotic non-destructive testing,” Ind. Robot, Int. J., vol. 37, no. 5, pp. 102–116, 2010. [6] R. Bogue, “The role of robotics in non-destructive testing,” Ind. Robot, Int. J., vol. 27, no. 5, pp. 421–426, 2011. [7] J. Shang et al., “Design of a climbing robot for inspecting aircraft wings and fuselage,” Ind. Robot, Int. J., vol. 34, no. 6, p. 495, 2007. [8] G. Dobie, R. Summan, S. G. Pierce, W. Galbraith, and G. Hayward, “A noncontact ultrasonic platform for structural inspection,” IEEE Sensors J., vol. 11, no. 10, pp. 2458–2468, Oct. 2011. [9] A. Ahrary et al., “A study of an autonomous mobile robot for a sewer inspection system,” Artif. Life Robot.. vol. 11, no. 1, pp. 23–27. 2007. [10] P. Louviot, A. Tachattahte, and D. Garnier, “Robotised UT transmission NDT of composite complex shaped parts,” in Proc. 4th Int. Symp. NDT Aerospace, Berlin, Germany, 2012, pp. 1–8. [11] Erhardt+Abt. (2016). HeiDetect FlexCT, Heitec. [Online]. Available: http://www.erhardt-abt.de/en/heidetect-flexct.html [12] D. Mery, Computer Vision for X-Ray Testing. Cham, Switzerland: Springer, 2015. [13] R. Prakash, Nondestructive Testing Techniques. Tunbridge Wells, U.K.: New Age Science, 2009. [14] A. W. Davis, T. N. Claytor, and M. J. Sheats, “High-speed data acquisition for three-dimensional X-ray and neutron computed tomography,” in Proc. 44th Annu. SPIE Meeting, Denver, CO, USA, 1999, p. 246. [15] D. R. Janecky et al., “WANTO LANL site report,” WANTO, Livermore, CA, USA, Tech. Rep. LA-UR-11-006214, 2011. This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. 12 [16] UniWest. (2011). Uni-Versal Robotic Eddy Current Test Machine. [Online]. Available: http://www.uniwest.com/Uni-Versal-Robotic-EddyCurrent-Test-Machine-P14.aspx [17] D. W. Bosserman. (Aug. 2007). The Future of NDT: Radiography Meets Robotics Quality Magazine. [Online]. Available: http://www.qualitymag.com/articles/84931-the-future-of-ndtradiography-meets-robotics [18] S. Niekum et al., “Learning grounded finite-state representations from unstructured demonstrations,” Int. J. Robot. Res., vol. 34, no. 2, pp. 131–157, 2015. [19] Yaskawa. (2012). SIA5D. [Online]. Available: http://www.motoman. com/datasheets/SIA5D.pdf [20] Robotiq. (2012). Adaptive Robot Gripper 3-Finger. [Online]. Available: http://robotiq.com/products/industrial-robot-hand/ [21] Robotiq. (2015). Adaptive Robot Gripper 2-Finger 85. [Online]. Available: http://robotiq.com/products/industrial-robot-gripper/ [22] J. J. Craig, Introduction to Robotics: Mechanics and Control, 3rd ed. Englewood Cliffs, NJ, USA: Prentice-Hall, 2004. [23] PaxScan 2520D/CL Amorphous Silicon Digital X-Ray Imager, V. I. Products, Palo Alto, CA, USA, 2011. [24] I. I. Gnezdilov et al., “Optimization of the neutron detector design based on the 6LiF/ZnS(Ag) scintillation screens for the GAMMA400 space observatory,” in Proc. Conf. Fund. Res. Part. Phys., 2015, pp. 199–205. [25] ROS. (2015). Robots Using ROS. [Online]. Available: http://wiki.ros. org/Robots [26] M. Quigley et al., “ROS: An open-source robot operating system,” ICRA Workshop Open Source Softw., vol. 3, nos. 2–3, pp. 1–5, 2009. [27] D. B. Pelowitz, “MCNPX user’s manual Ver. 2.7.0,” Los Alamos Nat. Lab., Los Alamos, NM, USA, Tech. Rep. LA-CP-11-00438, 2011. [28] L. P. Houssay, “Robotics and radiation hardening in the nuclear industry,” Ph.D. dissertation, Dept. Nuclear Eng., Univ. Florida, Gainesville, FL, USA, 2000. [29] J. Hashem et al., “Theoretical neutron damage calculations in industrial robotic manipulators used for non-destructive imaging applications,” Prog. Nucl. Energy., vol. 94, pp. 71–79. Sep. 2017. [30] J. Hashem, “Automating x-ray and neutron imaging applications with flexible automation,” Ph.D. dissertation, Dept. Mech. Eng., Univ. Texas Austin, Austin, TX, USA, 2015. [31] H. Heinisch, L. R. Greenwood, W. J. Weber, and R. E. Williford, “Displacement damage in silicon carbide irradiated in fission reactors,” J. Nucl. Mater., vol. 327, nos. 2–3, pp. 175–181, May 2004. [32] J. Hashem, J. Hunter, and M. Pryor, “Automating X-ray and neutron nondestructive testing applications,” in Proc. ANS Winter Meeting, vol. 109. Washington, DC, USA, 2013, pp. 517–520. [33] Manipulating Industrial Robots—Performance Criteria and Related Test Methods, document ISO 9283, International Standards Organization, 1998. [34] Idaho National Laboratory. (Nov. 23, 2015). MOOSE Simulation Environment, INL Fact Sheets. [Online]. Available: http://www4vip.inl.gov/ research/moose-applications/ [35] MATLAB, Image Processing Toolbox User’s Guide, MathWorks, Natick, MA, USA, 2015. [36] E. S. Jimenez, L. J. Orr, and K. R. Thompson, “High performance graphics processor based computed tomography reconstruction algorithms for nuclear and other large scale applications,” Sandia Nat. Lab., Albuquerque, NM, USA, Sandia Rep. SAND2013-8059, 2013. [37] Maglite. (2014). Maglite Solitaire LED 1-Cell AAA Flashlight, Maglite. [Online]. Available: http://robotiq.com/products/industrial-robot-hand/ [38] P. M. Silverman, C. J. Cooper, D. I. Weltman, and R. K. Zeman, “Helical CT: Practical considerations and potential pitfalls,” Radiographics, vol. 15, no. 1, pp. 25–36, 1995. [39] Newport Corporation. (2015). Newport, Compact Rotation Stage. [Online]. Available: http://search.newport.com/?x2=sku&q2=PR50CC [40] Radiation Safety Manual for Use of Radioactive Materials, ALARA Program, Washington Univ. St. Louis, St. Louis, MO, USA, 2014. [41] J. A. Gordon et al., “Millimeter-wave near-field measurements using coordinated robotics,” IEEE Trans. Antennas Propag., vol. 63, no. 12, pp. 5351–5362, Dec. 2015. [42] S. I. Chitta, A. Sucan, and S. Cousins, “MoveIt! [ROS topics],” IEEE Robot. Autom. Mag., vol. 19, no. 1, pp. 18–19, Mar. 2012. [43] MATLAB, Robotics System Toolbox, MathWorks, Natick, MA, USA, 2016. IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING Joseph A. Hashem received the Ph.D. degree in mechanical engineering from the University of Texas at Austin, Austin, TX, USA, in 2015, where he focused on robotics and nuclear engineering. From 2010 to 2015, he was a Research Assistant with the Nuclear Robotics Group, The University of Texas at Austin, where he designed dynamicallydeployed robotic systems for use in confined, hazardous environments and also taught and coordinated the Neutron Shielding Laboratory for both undergraduate and graduate students for a health physics laboratory course. He is a R&D Engineer with the Nondestructive Testing and Evaluation Group, Los Alamos National Laboratory, Los Alamos, NM, USA. Dr. Hashem received the Sally Blum Memorial Prize for Excellence in Design in Mechanical Engineering and the Hamilton Undergraduate Research Scholar from Southern Methodist University. He was a recipient of the Best Paper Award at the Third International Topical Meeting on Robotics and Remote Systems. Mitch Pryor is a Research Scientist with the University of Texas at Austin, Austin, TX, USA, where he co-founded the Nuclear Robotics Group (NRG). The NRG develops assistive and automated technologies for use in hazardous environments, including D&D sites, nuclear material storage facilities, and for glovebox manufacturing. Sheldon Landsberger is a Professor in the Nuclear and Radiation Engineering technical area in the Mechanical Engineering Department. He has served on the faculty of the Cockrell School of Engineering since 1997. He has published more than 220 peerreviewed papers and more than 160 conference proceedings mainly in nuclear analytical measurements and their applications in nuclear forensics, natural radioactivity, environmental monitoring of trace and heavy metals, and use of radiation detectors on robots and other mobile systems. James Hunter is a Research and Development Scientist at the Los Alamos National Laboratory, Los Alamos, New Mexico, USA, where he is a member of the Non-Destructive Testing Group and is team leader for the special engineering team. Since starting as a student at LANL 17 years ago, he has written much of LANL’s tomography reconstruction code and has worked with a range of X-ray systems from small micro-focus cabinets to a 6-20 MeV Microtron based system. His current work involves both production and R&D imaging on a huge range of components with both lab-based X-ray systems as well as synchrotron beam lines and neutron imaging facilities. His current work involves both production and R&D imaging on a huge range of components with both lab-based X-ray systems and synchrotron beam lines and neutron imaging facilities. David R. Janecky received the A.B. degree in geology from the University of California at Berkeley, Berkeley, CA, USA, in 1975, and the Ph.D. degree in geoscience from the University of Minnesota, Minneapolis, MN, USA, in 1982. Since 1984, he has been a Scientist and the Manager at the Los Alamos National Laboratory, with research interests in dynamic geoscience, industrial, and engineered systems, including nondestructive testing and evaluation, environmental remediation, global security, and support for the nuclear weapons stockpile.