Minimum-Invasive Self-Evolving Diagnostic Systems: An ultimate component of the Factories of the Future

The proposed project is motivated by the widespread need for condition monitoring and diagnostics of production processes in factories of the future. By no doubt, knowing the condition of production assets is vital for the operation of the entire production process. A faulty or abnormal condition of production assets may lead to inadequate product quality, decreased productivity, increased hazard, excess energy consumption and increased environmental impact. Detection and alarming of faults and degradations in early stage, well before they cause serious impacts to the production process, is therefore of great importance. This is recognized clearly and emphasized by Smart factory and Industry 4.0 concepts. The research and development of process diagnostics is a mature field. The resulting methods are mostly based on mathematical models of processes of various forms. The diagnostic is performed by comparing the output variables of the production process and its model. However, a lot of effort of experimentation, analysis, development and fine-tuning is needed to build the mathematical model and this leads to high costs. Moreover, the resulting model is valid only for a particular type of process. Modelling can only be performed by highly trained experts skilled in data analysis, modelling and understanding the background of the process which is a subject of diagnostics. All this represents a bottleneck and prevents a wide and smooth transition of diagnostic methods from theory to practice. The proposed project has the ambition to accelerate the use of diagnostic methods in industrial practice by overcoming the mentioned obstacles. This will be achieved by developing a new methodology of a minimum-invasive and self-evolving diagnostic system that could be applied widely and with only moderate effort. The project will provide a new and general enough diagnostic methodology that has strong potential to support different kinds of production processes, predominantly in the class of repetitive production processes, which include manufacturing, assembly and also batch process industry. During operation, processes of this kind generate signals with repeating elementary patterns. Faults or degradations of the process components cause changes in the elementary pattern. The main idea is to develop a diagnostic method that will sample the process signals and autonomously identify the elementary repeating patterns, decompose them into segments, characterize their features and determine nominal values and distributions of particular features. This represents the ‘’learning mode’’, which has to be active for some time until reliable information about the process is collected. After learning mode, the system is switched to ‘’exploitation mode’’ where diagnostics and condition monitoring starts. In this mode, the system continuously samples the signals, identifies the generated elementary repeating patterns and analyse them. Differences between the nominal and actual values of features of elementary patterns and their sub-elements indicate the presence of fault and degradations of a particular item of equipment. The proposed diagnostic system is considered minimum-invasive since it requires no modifications of existing process control systems, no significant modifications of electric wiring and no additional sensors. The diagnostic system is self-evolving, which means that no manual modelling by experts is required, diagnostic system collects all the necessary information automatically, during the learning phase and also later, during the exploration mode. The potential impact of the proposed project is enormous since the class of repetitive discrete production processes is very wide and numerous. Thanks to non-invasive property and self-evolving ability, the system has the potential to be smoothly applied to many production processes in Slovenia and abroad.