Specification of an intelligent simulation-based real time control architecture: Application to truck control system

Abstract

The paper presents an implemented architecture of an intelligent simulation-based real-time control (SRTC) system for industrial applications. The proposed SRTC uses a trajectory tracking strategy inspired from the model-based predictive control approach. Dynamic control law based on the closed-loop feedback correction is embedded. A computer implementation of this control scheme and experiments are conducted for real-time truck dispatching on a surface mine transportation system. Results showed the capability of the SRTC to generate efficient real-time truck dispatching orders at each 120 s. Simulation results demonstrate that managing trucks with such dynamic control law improves productivity. This improvement is reached when the transportation system is under steady as well as transient states conditions. The proposed SRTC makes use of the intelligent metaheuristic optimization search even under tight timeliness constraints.

Introduction

Over the last decades the need is growing for more dynamic and efficient control software to address productivity challenges in complex systems. A special emphasis is made to embed real-time control in complex manufacturing, transportation and service systems. For complex systems evolving in the highly stochastic environment the resort to real-time (on-line) control becomes unavoidable. The objective is to use a control mechanism which operates concurrently with the real world system and embeds a dynamic control law. The control law must be regularly updated to account for the current state of the system as it evolves in real-time. This infers a more efficient control as there is no more need to compute off-line decision/control laws based on simplistic assumptions such as the deterministic behavior or unrealistic modeling of the system uncertainty.

As an example of recent attempt to embed such real-time control in the real-world industrial context job assignment in the semiconductor back-end scheduling problem [1], real-time vehicle/truck dispatch in transport system under highly outside variable environment [2], real-time agent schedule adjustment in call centers to face the highly stochastic call arrival process [3], etc. In such applications, it was proved that in order to reach better performances the control law must be dynamically specified, under short term control horizon within the range of few minutes, to take into account the inherent system state variation.

This paper provides a functional architecture for the computer implementation in industrial applications of such intelligent simulation-based real-time control (herein after called the SRTC) software. The use of the discrete event simulation rather than the classical analytical modeling approach is chosen in order to offer more accurate and high-fidelity predictive model to the control system. The structural design and the behavior of the proposed SRTC architecture are built upon the object-oriented modeling methodology based on the Unified Modeling Language (UML) [4]. A specification phase exhibits how fundamental characteristics of real-time applications (timeliness, concurrency and reactiveness) are designed and implemented in the SRTC software. These characteristics are embedded in the architecture to ensure that the control law is both accurate and timely.

According to [5], [6], [7], even if the SRTC has the potential to lead to a more efficient control, its embedment in real-world application remains sporadic. In [5], authors point out the lack of formal modeling of the simulator generation, under real-time constraints, and the related online data interaction with the physical shop floors. Mirdamadi et al. [6] show that most of the SRTC reported works describe particular issues related specifically to shop floor control applications. They also reveal that most of these SRTC operates in an open-loop control manner due to the complex implementation of online data interaction. In [7], Iassinovski et al. associate the scarcity of the SRTC to the lack of specification methodology leading to a computer software architecture integrating the simulation model with the control system. Thus, the purpose of this work is to give more insights into the formal modeling and the computer implementation of the promising SRTC in industrial applications.

The herein developed SRTC architecture is based on notions of sensor feedback and reference trajectory, inspired from the successful method in control theory of model-based predictive control [8]. This SRTC system evolves concurrently, at the operational level, and deals with continuous variability by proposing a dynamic control law based on a closed-loop feedback correction. The controller triggers a new simulation-based optimization process based on the updated system state. For this purpose, an internal simulation model is used to predict the system's output over the prediction horizon. The objective of this optimization process is to find a control law for the next control horizon that minimizes a cost function defined as the gap between the actual reached performance delivered by the sensors and a reference trajectory. This reference trajectory is computed off-line to respond to a production plan or to managers’ performance targets.

Another feature of the proposed architecture is the use of a metaheuristics, the Simulated Annealing, during the simulation-based optimization process to compute the near-optimal control law. Researchers agree that the intelligent searching method behind the metaheuristic optimization techniques made them more powerful than the classical ones [9]. However, according to [10], [11], their use at this real-time control stage is discarded due to their integration complexity and to the long simulation run-time. Section 3.2 provides explanation on using multiprocessing technique to alleviate the integration complexity. Section 4.2 shows that the simulation run-time could substantially be decreased by developing a specific simulation model designed for the control purpose.

In this work, a proof of feasibility of the architecture is carried out through a prototype implementation of truck control software in surface mine transportation system. Conducted experiments show the capability of the designed software to resolve, in real-time, truck dispatching problems in a stochastic surface mine transportation system. With the proposed software a dynamic control law, herein expressed as a vector of trucks that are dispatched to pick-up stations, can be generated at each control horizon of 120 s. This leads to a more efficient truck management under the dynamic and unpredictable non-linear traffic behavior in the transportation system. Furthermore, as the SRTC is implemented in a concurrent computing environment, its interaction with the real-world system is continuously maintained and the occurrence of external perturbations is always tracked with the closed-loop sensors feedback.

This paper is organized as follows. Section 2 reviews and discusses existing SRTC architectures. Section 3 deals with the specification phase of the proposed SRTC. Section 4 details the implementation and gives an overview of the proposed architecture. Section 5 includes simulation experiments and discusses results. Finally, Section 6 summarizes the findings and the main conclusions.