Available online at www.sciencedirect.

com

ScienceDirect
Available online at www.sciencedirect.com
Procedia Computer Science 00 (2018) 000–000

ScienceDirect www.elsevier.com/locate/procedia

Procedia Computer Science 120 (2017) 411–416

9th
9thInternational
InternationalConference
Conferenceon
onTheory
Theoryand
andApplication
Applicationof
ofSoft
SoftComputing,
Computing,Computing
Computingwith
with
Words and Perception, ICSCCW 2017, 24-25
22-23 August
August 2017,
2017, Budapest,
Budapest, Hungary
Hungary

Soft computing models in an intellectual open-pit mines transport

control system
Temkin Igor*, Deryabin Sergey, Konov Ilya
National university of science and technologies MISiS, 119049, Moscow, Leninsky Ave., 4, Russia

Abstract

This article focuses on the development of control mechanisms of open pit mine robotized transport system. Method for the state
of the technological roads identification with the use of fuzzy inference mechanisms is discussed.
The structure of the control system is presented. The possibilities of the use of telemetry data for a wide range of important
technological problems, which, anyway, are reduced to problems of interpretation of the data, object identification, forecasting of
parameters and the interaction of robotic agents control are discussed. The examples of the formal setting of some of these
problems which based on the soft computing are presented. Considered in detail the problem of identification of the open pit
mine roadway. The procedure for organizing and conducting dump truck onboard experiments is described. The features of the
main control parameters are analyzed. The mechanism of using the telemetry data processing in common with fuzzy inference
tools is presented. The main stages of the experiment telemetry data processing are described.

© 2018 The Authors. Published by Elsevier B.V.

Peer-review under responsibility of the scientific committee of the 9th International Conference on Theory and application of
Soft Computing, Computing with Words and Perception.

Keywords: open pit mine; mining and transport system robotized objects; identification and evaluation of the mining road state; telemetry data;
fuzzy inference; learning mechanisms

Nowadays, creation of pilotless technologies for developing quarries and various mineral and ore deposits is a
worldwide trend. The most important part of this technology is transporting mineral and ore mass via robotized or
completely autonomous mining transportation. Many leading companies such as Komatsu (Japan), Caterpillar

* Corresponding author. Tel.: +7-499-230-2471

E-mail address: igortemkin@yandex.ru

1877-0509 © 2018 The Authors. Published by Elsevier B.V.

Peer-review under responsibility of the scientific committee of the 9th International Conference on Theory and application of
Soft Computing, Computing with Words and Perception.

1877-0509 © 2018 The Authors. Published by Elsevier B.V.

Peer-review under responsibility of the scientific committee of the 9th International Conference on Theory and application of Soft
Computing, Computing with Words and Perception.
10.1016/j.procs.2017.11.257
412 Temkin Igor et al. / Procedia Computer Science 120 (2017) 411–416
2 Temkin Igor et al. / Procedia Computer Science 00 (2018) 000–000

(USA), Euclid-Hitachi (Japan), that produce machinery and transport solutions for mining industry and conduct
active research in heavy trucks and excavators for open pit mining (Klebanov, 2015; Temkin, Klebanov, 2014).
One of the more perspective directions of research is developing effective intellectual control systems, based on
technical possibilities of robotized machinery and modern methods of data manipulation. A simplified scheme of a
mining technological complex is represented in Figure 1. The main elements maintaining the mining-transport
complex (MTC) are:

 on board computing system, installed inside quarry tippers, that maintain multifunctional control (collecting,
computing an transferring navigational and diagnostics data to the operations center) and remote control of the
MTC;
 GLONASS\GPS navigation complex, that are installed on every single mobile unit and allow it’s positioning
with up to 1 – 1,5m precision, as well as various sensors that monitor unit’s condition;
 software complex that allows for centralized remote control;
 wireless data transmission systems: Ultra Short wave connections, tracking systems, mobile connection systems,
as well as broadband connections and wireless data transmission systems, such as Wi-Fi, WiMAX and MESH
systems that supply reliable radio coverage and high speed transfer, sufficient for real time video feedback and
any diagnostic information (Brown, C., 2012; Temkin, Klebanov, 2014).

Fig. 1. Simplified MTC scheme

The concept of intellectual MTC control system is based on step by step expansion of system’s capabilities in
context of robotized machinery integration (prioritizing heavy trucks and excavators) and usage of modern
technologies of data analyzing and processing (geological, operating, technological, sensor and expert data). In such a
case many of the traditional MTC automated control systems objectives are non-relevant any more, others are
modified. These steps, characterizing system’s capabilities are sorted by control reactions speed and increasing
autonomousity of technological machinery and “pilotlessness” level of the control process:

 centralized supervisor control, based on operative planning models and mining-transport machinery;
 supervisor control based on sensory data of partially robotized MTC units;
 Temkin Igor et al. / Procedia Computer Science 120 (2017) 411–416 413
Temkin Igor et al./ Procedia Computer Science 00 (2018) 000–000 3

 complex functioning control of robotized units with wide variety of functions (diagnostics, interpretation,
identification, predicting);
 remote control of robotized agents;
 remote control and interaction of partially-autonomous agents.

Final step of creating an intellectual control system for all technological processes of MTC functioning, such as:

 a heavy truck following a path from load to unload point;

 loading a heavy truck with an excavator at a loading point;
 unloading a heavy truck at an unloading point;
 heavy truck movement from it’s parking spot to an objective assigned quarry track;
 heavy truck following directions.

The system is a complex of interacting intellectual agents, that have varying behavior models and basic parameters.
Intellectual agents is a complex of software modules that exchange input and output information (Temkin, Klebanov,
2014; Kammoun et al., 2014; Claes, Holvoet, Weyns, 2011).
Several agent classes are defined:

 operator agent. Basically, a self-educating agent, capable of forming a situational model without human
interaction by collecting, accumulating and sorting various facts, events and situations from other agents and
interacting with them, capable of choosing the best strategy for fulfilling it’s objective;
 infrastructural agent units, that act as static technological environment elements: loading bays, roads, traffic
intersections, etc. Each of this agents is described by an array of parameters, and other agent’s interaction with
them is followed by a corresponding set of rules;
 reflective autonomous agents are system elements whose behavior is based on environmental conditions and
information from other agents. All robotized units, such as heavy trucks, excavators, conveyor belts, maintenance
vehicles are considered this class.

Autonomous functioning and agents interaction is based on two baseline models: cognitive, that are used for
describing agent behavior and strategy selection conditions; and processing, that are used for agent condition and
technological situations interpretation, agent planning optimization, identification as well as data base parameters
optimization (Temkin, Kulyanitsa, Kubrin, 2015; Bingnan Jiang, Yunsi Fei, 2016).
The system’s functioning is mainly dependent on information data base that is formed using every control system
set up in an open-pit mine and all of the available telemetric information. The most developed MTC element nowadays
is heavy truck on-board computing devices. Constant moving of select heavy trucks on an industrial site allows for data
collection of not only the unit itself, but of its surrounding environment as well. Depending on the current objective the
data collected can be processed by on-board computers in real-time and transferred to remote servers for more detailed
analyses.
The telemetric data collected by the mobile units’ on-board computers can be sorted into several segments
considering its role, that is represented by varying parameters in the control system:

 data regarding mobile unit’s positioning;

 parameters, characterizing mobile object’s functioning data (fuel levels, load, movement speed, suspension load,
etc.);
 integral indicators, reacting to precise unit’s movements – unit’s vector acceleration sensors. [α(x), α(y), α(z),
α(x, y), α(x, z), α(y, z), α(x, y, z)];
 parameters, describing unit’s position relative to static units in the environment;

Let’s see how system’s modules interact with the data base while managing the movement of mobile agent i.
The variables would be:
414 Temkin Igor et al. / Procedia Computer Science 120 (2017) 411–416
4 Temkin Igor et al. / Procedia Computer Science 00 (2018) 000–000

~
 X[k]  {x 1 , x 2 , x 3 } – agent’s coordinates in time point k;
~
 U[k]  {u 1 , u 2 , u 3 } – vector of parameters registered by the on-board computer;
 γ[k]– parameter that describes the agent’s condition;
~
 V [ k ], Vˆ [ k  1] – current and predicted agent’s velocity;
 S[k]={S1, S2, S3, S4}– vector of parameters that allow to evaluate current transport and technological situation;
 Tˆ [ k  1] – route selection at the k+1 point in time. A typical situation would be T{x[k]}=…=T{x[k+t0]}, t0 –
estimated time of arrival to a key point of the route;

Every mobile agent is also collecting data for a constantly updating online database of site’s track quality. Therefore
Dx[k] is a parameter that describes the track’s condition at the agent’s current coordinates at the k point in time.
A scheme of software modules interaction in agent’s moving processing is shown in Figure 2.

Fig. 2 – Scheme of an intellectual system’s primary functional nodes interaction

Considering that a major segment of the data base (exc. the technological regulations and mining facility safety
norms) consists of fuzzy logic and soft computing based models and arrangements, a few more variables need to be
stated:

 Lu1=Z1,Lu2=Z2,…,Lup=Zp (p≤n) – linguistic variables array;

 φ1((Zi), φ2((Zi),…, φs((Zi) – functions, stated in both triangular and exponential form;
 Z i1 ,..., Z is – a variety of terms, based on the following condition: ximin  x  ximax ;

Thus considering, the term’s membership function is dependent on two or three parameters, while the variety of
parameters that need to evaluated shall be designated as A .
~ *
The fuzzy inference algorithm in this case shall be designated as the following: F1 : {R f , A, xi }  yi , where
*
yi is the fuzzy inference algorithm output.
The agent’s moving procedure thus can be described in the following way:

F1 : {RF , A , ~ x [ k ], u~[ k ]}  D [ k ] ;
*

 F2 : {~ x [ k ], u~[ k ]}   [ k ] ;
F3 : {S [ k ], D [ k ]}  Vˆ [ k  t ] ;
* *

F4 : {~ x [ k ], D [ k ],  [ k ]Vˆ [ k  t ]}  xˆ[ k  t ] ;
*

* *
 F5 : { xˆ[ k  1], S [ k ], D [ k ]}  T [ k  t ] , where Δt is the minimal interval between control requests.

F1-F5 realize several different processing models that unite the hard and soft calculation mechanisms: adaptive
neural networks, traditional backpropogation neural networks, evolutional optimization methods and optimal flow
distribution algorithms.
As was mentioned before, monitoring of the track’s condition is essential. Let’s consider it in further detail.
The data base is updated with not only real-time mobile agent positioning, but also other sensor data. For
 Temkin Igor et al. / Procedia Computer Science 120 (2017) 411–416 415
Temkin Igor et al./ Procedia Computer Science 00 (2018) 000–000 5

example, every heavy truck transmits data about the strain of it’s wheel suspension: Ufr(t), Ufl(t), Ubr(t), Ubl(t).
For creation a track condition diagnostic model some functions, comfortable for evaluation and expert analyses,
were used. They are considered as linguistic variables:

 Z1(t)=[Ufl(t)+ Ubl(t)]- [Ufr(t)+ Ubr(t)] – «Chassis roll»;

 Z2(t)=[Ufl(t)+ Ufr(t)]- [Ubl(t)+ Ubr(t)] – «Chassis pitch»;
 Z3(t)=[Ufl(t)+ Ubr(t)]- [Ufr(t)+ Ubl(t)] – «diagonal distortion»;
~ ~
 Z 4 (t )  V (t )  V (t  t ) – «velocity variation».

Figure 3 represents a fragment of listed variables realization for one of the mobile agents.

Fig. 3 – Fragment of realization of variable suspension load and velocity

Track condition evaluation was solved in two variants:

 Calculating a rating score of select road segments condition;

 Recognition of a specific track anomaly (“pit”, “hummock”, “pit combination”);

Linguistic variable D, “track condition” – is based on five terms: {“very poor”, “poor”, “satisfactory”, “good”,
“very good”}.
This way each of the listed linguistic variables was stated on a universal term array {NB, N, E, P, PB}. As the
allegiance functions triangular φE(Z,α,β,δ) and Bell’s functions φE(Z,α,β) were used.
Two type of rules were researched when forming a rule base:

 If Z1=N and Z2=E and Z3=P and Z4=NB then D=N;

 If Z1=N and Z2=E and Z3=P and Z4=NB then D= Ci, i =1,q (q – quantity of specified track disturbance classes);

Tsukamoto and singleton schemes were used as an inference mechanism, while the parameters that form
antecedent part of the rules were evaluated at the learning stage with use of traditional data
processing gradient and genetic algorithm technologies in S ( Dˆ  D* ) 2 order to minimize the criteria, where s
j 1 j j *
is overall amount of evaluated tracks, D̂ j – expert evaluation, D j - fuzzy system’s
416 Temkin Igor et al. / Procedia Computer Science 120 (2017) 411–416
6 Temkin Igor et al. / Procedia Computer Science 00 (2018) 000–000

inference.
The described scheme of soft computing based road condition evaluation is implemented in the following steps:

 Forming a set of rules based on analyzing a significant amount of varying heavy truck routes (80 to 90
fragments) in order to classify track anomalies. 8 classes were specified, of which 7 are directly connected with
the actual road condition, and 1 with a complicated transporting condition – “sudden stop”.
~ ~
 Testing heavy truck relocations in order to range the quality of certain road segments. { X (t ), U (t )} parameters are
specified at this stage.
 Array of rules optimization (reduction of rules quantity, from potential 625 to around 30-40 entries) and
preliminary set up of agent membership functions.
 Tuning real time agent membership functions parameter identification algorithm using imitation modelling.

During testing of researched models on new for the mobile agent routes some efficiency ratings were acquired.
1. The experiments were conducted on 0.5 to 4.2 km routes; the amount of track anomalies varied from 1 to 7
(anomaly classes amount Ci=4). Rating analyses of track condition was conducted in 45 zones. As a result, the worst
classification precision rating equaled 75%:
~
1 1, Di*  Ci
i 1 i *100%; i  0, D*  C~ ; N  7 .
N
K1 
N  i i

Modelled rating scores were significantly higher, ≥ 87%:

1 1, Di*  Dˆ i
i 1 i *100%; i  
N
K2  ; N  45 .
N 0, Di*  Dˆ i
2. This leads to a conclusion that the second variant of route condition evaluation is preferred.
3. Further researching before integrating the module into real practice will be aimed at investigation of the
efficiency of all the above mentioned learning algorithms with real-time fuzzy logic based identification rules data
base. Nevertheless, the question of evaluating the track condition by using multiple different agents data is still of an
issue.
The study was carried out at the expense of a grant from the Russian Foundation for Basic Research (RFBR)
(project № 16-07-01197)

References

Bingnan Jiang, Yunsi Fei, 2016. Vehicle Speed Prediction by Two-Level Data Driven Models in Vehicular Networks. IEEE Transactions on
Intelligent Transportation Systems, November.
Brown, C., 2012. Autonomous technology in mining. Engineering and mining journal, January.
Claes, R., Holvoet, T., Weyns, D., 2011. A decentralized approach for anticipatory vehicle routing using delegate multiagent systems. IEEE
Transactions on Intelligent Transportation Systems, pp 364–373.
Kammoun, H.M., Kallel, I., Casillas J., Abraham, A., Alimi, A.M., 2014. Adapt-Traf: An adaptive multiagent road traffic management system
based on hybrid ant-hierarchical fuzzy model. Transportation Research С42, pp. 147-167
Klebanov, А.F., 2015. Informacionjie systemi gornogo proizvodstva i osnovnie napravlenija razvirija avtomatizacii otkritih gornih rabot. Gornaja
promishlennost 2 (120).
Temkin, I.O., Klebanov, D.A., 2014. Intellektualnie systemi upravlenija gorno-transportnymi kompleksami: sovremennoe sostojanie, zadachi I
mechanismi reshenija. Gorny informacionno-analiticheskij bjuletten, Megdunarodnaja nauchno-practicheskaja konferencija «Nedelja
gornyaka», Specialnii vipusk.
Temkin, I.O., Kulyanitsa, A.L., Kubrin, S.S., 2015. Application of intellectual system for robotic coal plough machine. “Miner’s week – 2015:
reports of the XXIII international scientific symposium”.