Thesis of Behnam Einabadi
Start date: 01/10/2019
End date: 01/10/2022
Advisor: Armand Baboli
In the highly competitive global marketplace, manufacturing companies must adopt a strategy to expand their product offering with added functional diversity to capture maximum market share while ensuring quality. This requires flexible and proactive production systems which are named as Industry 4.0, Intelligent Manufacturing Systems (IMS), Cyber-Physical Production Systems (CPPS), Smart Factory, etc. is today the most relevant and effective response to current challenges and is the only source for sustainable competitiveness. These systems are equally adaptable for service system (for example hospital).
As a result of technological disruptions in ICT, these systems provide access to big-data resulting from sensors of production and transportation equipment as well as several information systems. Consequently, based on the resulting information and knowledge from predictive models which are build using AI and machine learning techniques, potential benefits are collaborative decision making via IoT: Internet of things (neither centralized nor decentralized) and individualized mass production (one-of-a-kind manufacturing and assembly) for end to end global optimization (systematic approach). The decisions taken as measures for optimization can then be readjusted upon integrating real-time data and information.
Because of usage of high-level advanced production and communication technologies, these systems are very complex and it is difficult to manage them efficiently, effectively and optimally. one of the main challenges in Industry 4.0 is the maintenance and availability of the production system and equipment. The solution to this challenge requires optimal production plan and maintenance activities, simultaneously. As a result, global optimization is more difficult due to the fact that these systems are structured on multiple types of equipment where components can have strong dependencies (functional, technological, and economical). Therefore, in this thesis, we focus on the organization and optimization of predictive maintenance activities which not only directly impact the availability of production system capacities but also the potential desirable sustainable quality behavior.
Most of the traditional methods use only historical data of the equipment and the failures and propose preventive maintenance based on Mean Time Between Failure (MTBF) and some time to readjust it with new information. More recent approaches propose the use of sensors data for monitoring degradation of parts and components, referred as condition monitoring, to evaluate the potential failures and plan anticipative actions to avoid future failures. However, degradation modelling methodologies and identification of the best time for maintenance upon estimating remaining useful life (RUL) by taking into account not only functional, technological correlation and dependencies, but also the correlation and dependencies between production planning activities and replenishment activities of spare parts, remains a major scientific challenge. There is a rise of interest in the data-driven approaches for predictive maintenance.
The RUL based methodologies for degradation modelling are focused on components and are primarily based on physical models which include material properties. All such methods are demonstrated in the literature using historical data; however, this becomes more complex where predictive decisions are to be adjusted based on real-time data and information collected from the production system. Ideally, maintenance operations cost and production equipment expert’s knowledge, if integrated, will result in global optimization where unscheduled equipment breakdown may also be reduced while avoiding occurrence of known failures and ensuring sustainable production quality [25, 26,27].
The main objective of this thesis is to develop models and methods for optimization of dynamic predictive maintenance for production system exploiting the concept of industry 4.0. To achieve this objective, following three scientific contributions are envisaged.
Firstly, it is necessary to define the remaining useful life (RUL) of not only component but also at the parts level. In this respect, we focus on using the correlation and dependencies between components and parts which are modelled in parent-child relation within the equipment assemblies, by exploiting the sensors data (vibration, temperature, energy consumption, air pressure and noise level) and information collected from other sources (ERP, WMS, MES, etc). The novelty of this approach is to take into account not only the components but also the parts in parent-child relation, integrating big with multisource data for correlation and dependencies, to develop predictive models based on AI and machine learning techniques (neural networks, Bayesian network, machine learning, deep learning, artificial intelligence, etc.) in real-time environment.
The second contribution principally focuses on the maintenance tasks planning and scheduling. In this step, all parts, components, and equipment subjected to potential maintenance, resulting from first step, are further considered for scheduling based on cost (parts replenishment, inventory, obsolesce; and maintenance fix [setup cost], variable cost [operating time for changing spear parts], production cost, over and under quality cost, etc.) availability of spare parts and technical resources via a multi-objective mathematical model and appropriate resolution algorithms to identify optimal time for maintenance task planning. The model objectives include cost minimization, (or profit maximization) as well as service-level optimization. In this step, traditional operation research, dynamic programming or goal programming will be used. Moreover, the possibility to transform the results of this mathematical optimization into decisions will also be studied. The aim of this transformation is to reduce the time from the optimization engine’s task planning and scheduling results until final decisions are locked for execution in real-time settings.
The next step concerns the rules from human expertise. The objective of this step is to study the possibility of improving maintenance strategy by consideration rules developed by maintenance experts from previous steps predictive decisions. These thumb rules will serve as the third optimization layer prior to the execution of tasks planned and scheduled for potential maintenance activities at components and parts levels. The main contribution of this step is a methodology to identify more accurate maintenance (CM, PM) tasks for execution in an optimal fashion based on Bayesian network.
All three steps will be tested using data collected from an automotive or aeronautic industry which are highly demanding not only in the offering of diverse functionality of products but also with sustainable quality for key competitiveness.