Engineering Headway Vol. 12

Abstract: This study presents an approach to improve cocoa harvesting using image recognition technology. Convolutional neural networks (CNN) along with the help of the Roboflow platform were used to analyze images of cocoa fruits and determine their maturity stage in an uncontrolled environment. The study focused mainly on the effectiveness of taking pictures at different distances, dividing the images into three categories (0.10m-0.30m, 0.30m-1.00m and 1.00m-3.00m) each trained with 400 images in order to evaluate the performance of each one in terms of its mAP, precision and recall. Subsequently, a fourth neural network was developed using all the data collected, with a total of 1,200 images used for training, among which 3,255 unripe cocoa fruits and 2,555 ripe cocoa fruits were found. The results obtained were a mAP of 92.3%, accuracy of 91.6% and recall of 85.1%, demonstrating the high accuracy of the model in the classification of cocoa fruits. This neural network has the potential to improve the quality of the final product by accurately determining the state of maturity of the cocoa, which is essential for the industry.

Abstract: This prototype of a CO2 sensor with Internet of Things (IoT) offers an affordable solution for measuring carbon dioxide (CO2) concentrations in parts per million (ppm). It utilizes specialized sensors to detect CO2 concentrations in the environment. The prototype combines multiple CO2 measurement sensors with an Arduino microcontroller to process the collected data and provide comprehensible ppm CO2 readings. Additionally, an ESP-32 has been incorporated to enable IoT connectivity, allowing for the transmission of CO2 readings to a cloud platform. This platform displays the most recent readings and maintains a brief history of previous measurements, providing real-time insights into CO2 conditions and a record for analysis. A noteworthy feature of this prototype is its buoy system, which enables operation in aquatic environments while minimizing the risk of submersion, ensuring that the sensor remains on the water’s surface without direct contact with the liquid.

Abstract: Acoustic testing is a technology that covers various machinery failure modes, including bearing and gear failures. This technology is superior to vibration analysis for gear and bearing condition monitoring. This paper aims to offer the maintenance world a critical technological advance by developing a web-based tool that, using pretrained convolutional neural networks and spectrograms, allows the diagnosis of gearboxes from recordings obtained with industrial acoustic testing tools. The resulting model is tested against human specialists to assess its actual world performance. A modified agile methodology was implemented to develop the research systematically. The type of approach is mixed since it has qualitative parts, such as specialists involved in obtaining the ultrasonic data and classifying them, and quantitative parts, such as validating the precision of the model based on established validation metrics. By using a pretrained model and then performing a fine-tuning with heterodyne ultrasound recordings from gearboxes in good and bad condition, a training accuracy of 93% was achieved. Then, tests were carried out to validate false positives and negatives in which it was possible to obtain 0% and 6.7% scores, respectively. This model was incorporated on a web platform to create the diagnostic tool whose input variable is the recording, and the output variables are its spectrogram, the prediction of whether it is in good or bad condition, and the probability of both possibilities.

Abstract: A memristor crossbar’s intrinsic device network dynamics can be harnessed to efficiently conduct radionuclide classification tasks by retrofitting the network with a peripheral CMOS-based architecture that has been structurally and functionally optimized for this classification task. However, the hardware implementation of this classification platform is limited by the physical characteristics of the memristor which has a finite number of states. This renders the employment of traditional neural network learning algorithms, where the weights are not limited to defined states, as an excessively complex task. Hence, this paper tests the limitations on weight resolution and its effect in classification precision when implementing a spiking locally competitive learning algorithm. Both linear and nonlinear weight distributions are examined. The algorithm’s local competitiveness is assessed for the specific application of radionuclide identification. The system is tested using spectra data obtained from the United States National Nuclear Data Center as the classification database dictionary. The platform’s accuracy is measured when test signals with 100, 10 and 1 signal-to-noise ratios are assessed. It has been shown that the system is highly effective for classifying radioisotopes with linear weight distribution even with high levels of noise present. A minor classification accuracy improvement was also observed for weight states distributions with a higher density of values in the low conductivity range. Therefore, it is concluded that a memristor-based radionuclide classifier should have at least 4 possible states mapping the algorithm’s synaptic weights.

Abstract: This project focuses on developing and implementing a robotic control system based on detecting signs and gestures using computer vision. The main goal was to create an intuitive and efficient interface for interacting with an OMRON Viper 650 industrial robot. To achieve this, computer vision technologies like Mediapipe and OpenCV were used to detect and recognize the user’s hands and fingers in real-time. The collected data was processed with a Python script and stored in a text file. Additionally, a program was developed in C# using OMRON’s ACE programming interface to extract data from the text file and send commands to the Viper 650 robot, enabling it to interpret the user’s gestures and perform actions accordingly. This project has successfully created an innovative solution that combines computer vision, programming, and industrial robotics to provide an intuitive and efficient control experience, opening up new possibilities in industrial and human-robot interaction applications.

Abstract: Technological progress introduces new ideas and methods that eventually end up being in another technology. One technology can not provide a solid ground for the future and, thus, technologies intervene with each other. A combination of technologies allows the achievement of greater application in different fields and areas and better performance and functionality. The Internet of Things is not a new idea, but due to the improvement of other technology can be used in different industries to achieve better optimization. Internet of Things can be found in different industries, but the focus of this work is in the field of robotics. Therefore, the following work will try to illustrate the use of technology in this field. A brief definition of the Internet of Things will be covered. The application of the technology in robotics will be outlined with its future perspectives. The advantages and drawbacks of the Internet of Things in the field of robotics will be discussed at the end.

Abstract: Soft robotics is an emerging subfield of robotics that studies the design and fabrication of automated systems composed of flexible materials. They present a potential solution to protect fragile objects from high stress induced by rigid grippers. This paper proposed a 3D printed soft pneumatic gripper adapted to precisely grasp delicate and fragile objects such as those encountered in the marine, electronic, and food industry. The gripper was based on the principle of a fluidic elastomer actuator and consisted of two soft TPU fingers and a rigid base with an Arduino-driven flexing sensor to measure the curvature of the fingers during grasping and a force sensor that enables precise measurement and adjustment of gripping force, ensuring the objects were held securely without damage. The design and fabrication were cost-efficient and engineered to not affect the continuous flexing of the soft fingers, addressing key challenges grasping with precision and efficiency. The relationship between the sensor outputs and pneumatic inputs were analyzed through graphs from conducted experimental tests.

Abstract: The present study reflected on the role of teachers of architects and civil engineers in the promotion of entrepreneurship as an institutional axis of higher education centers with the aim of identifying the entrepreneurial competencies they possess through a self-diagnosis based on a mixed approach through a survey questionnaire established after a literature review of the subject and considering the opinion of experts. resulting in the profiles of teachers, their knowledge, skills, social roles, character traits, values and motivations that they possess in their classrooms, as well as the learning strategies they use, character traits, values and motivations that they have in their classrooms, as well as the learning strategies that they suggest they use, in contrast to the theory that, It generalizes teachers in their role as facilitators of learning, but there is an opportunity to improve those skills such as communication, and it is important to provide spaces and activities for both teachers and students to socialize with professionals and companies that are currently part of the entrepreneurial ecosystem in their cities. The perceived limitations to achieve success in the entrepreneurial field were also made known and the interest of teachers in continuing to innovate and train to apply and master this knowledge was documented. It is suggested to venture into sustainable social entrepreneurship since it allows to address environmental and socioeconomic projects, promoting a strong ethic and responsible action in the face of various community problems.

Abstract: The purpose of the study was to evaluate the performance of neural networks as modern techniques to classify the risk of default against the traditional Logit statistical method, taking a Honduran bank as a case study. The data was obtained from its credit portfolio made up of 38,156 personal loans and 9 available characteristics, choosing the most representative independent variables to design a Multilayer Perceptron type base model and its Logit equivalent to which characteristics were added to analyze their impact on the classification of the dependent variable Default, leaving in the end a network with an input layer of 8 nodes, 4 hidden dense layers of 20 and 24 nodes, a central dropout layer and a node in the output layer as well as an equivalent logistic regression to compare the performance of both. The results with unbalanced data showed a superior performance of the networks, but when applying SMOTE oversampling, although there was no greater impact on the network, there was in the regressions, concluding that these learn to classify loan default better when the data subsets are balanced in the class of the response variable since its new results almost reached those of the neural network, which was finally chosen as the preferred model for its implementation with an accuracy of 99.16%, precision of 99.47%, sensitivity of 99.59%, specificity of 95.48 %, F1 score of 99.53% and ROC and PR curves with AUC of 98.68% and 97.69% respectively.