Article Preview
Top1. Introduction
The constant technological change in industrial systems and the continual rise in equipment maintenance and labor costs, raise enormous problems to manufacturing companies, which requires solutions within the global competitiveness context. A key problem lies in the complexity of the industrial process technology (Munirathinam & Ramadoss, 2016). Systems degradation phenomena combined with manufacturing defects have the effect of reducing the capacity of equipment or machinery in providing the services for which they were designed within a planned timeframe. The diversity of theses degradations and defects intensifies the problem and adds to the failures or breakdown a random dimension.
Numerous approaches and solutions have been proposed in the literature to deal with such phenomena and thus obtain the best compromises between the availability of equipment, production costs, product quality, and competitiveness. These approaches and solutions are found in the fields of maintenance, risk analysis, fault diagnosis systems (Wang, Ma, & Tu, 2011), or prognosis (Dong & Yang, 2008) of industrial systems.
Among the means currently used in manufacturing companies today, we can mention the statistical approaches to process control which consist in controlling production chains using performance indicators (Chang & Sun, 2004), and those for the detection and classification of defects which aim to monitor changes in equipment or machinery parameters in real-time. Thus, whether corrective, preventive, or predictive, the various maintenance strategies are mainly part of the solutions applied.
However, we note that some of these approaches, even combined, do not make it possible to control the high variability linked to production equipment introduced in a complex and uncertain industrial context.
In complex industrial systems, the current trend is to use networks instead of an analytical approach, as a more comprehensive modeling tool (Thamhain, 2013). Networks have a better potential for 1) analyzing equipment behavior and detecting the causes of failures or breakdown, 2) solving breakdown propagations (Hu, Zhang, Ma, & Liang, 2011) measuring the overall equipment performance.
The objective of this article is to use Bayes theory (Daly, Shen, & Aitken, 2011), particularly Dynamic Bayesian Networks (Khakzad, Khan, & Amyotte, 2013) applied to industrial control systems to predict process failures and prioritizing the risks and their causal structure. These systems rely primarily on Supervisory or Control of Data Acquisition (SCADA), an interconnexion of components, or a control system architecture, that uses computers, networked data communications and graphical user interfaces for high-level process supervisory management. The overall objective is therefore to help assessing and solving reliability problems of complex industrial systems (Li, Chen, Liu, Liu, & Dai, 2018).
Bayesian networks have become a reliable tool in Artificial Intelligence (AI) to model industrial systems' contingencies and exploit them in decision-making. More precisely, they make it possible to represent probabilities in a very compact manner on a computer, and they provide explanations and solutions to calculate those useful for decision-making efficiently.
Bayesian networks are compact graphical representations of large-dimensional probability distributions. They are equipped with very efficient inference algorithms to calculate probability distributions.
The relationship between AI and Bayesian networks is in the performance and flexibility, combining several aspects, including statistics, probabilities, decision support, and knowledge management of static and dynamic systems. The Bayesian approach offers good and qualitative modeling for complex industrial systems with dependencies and non-dependencies between multistate random variables.
The dynamic Bayesian approach for a complex industrial system is quite powerful. Engineers, scientists, researchers, and other experts have well-known reliability analysis tools like fault trees, reliability diagrams, and Failure Mode and Effects Analysis (FMEA) to know the system's state before deciding. However, artificial intelligence tools, such as Bayesian networks, can provide practical and innovative help in making decisions about the operation, maintenance, or reducing dangers/risks for industrial systems.