An Approach to Model-based Fault Diagnosis of Manufacturing Processes and Machines using Probabilistic Boolean Networks

Developing systems and methodologies capable of monitoring the condition and diagnosing multiple faults in industrial/manufacturing systems are topics of active and continuous research. In this paper, a fault diagnosis system inspired on the Probabilistic Boolean Networks (PBN) with Intervention model is suggested as a tool for diagnosing faults of a group of machines in a manufacturing process. The proposed approach considers the failure modes of the machines involved that are affecting the function and performance of the system. Firstly, the modes are identified and divided into two groups: faults and failures. The former implies detectable degradation of system function until the threshold for fault, which is eventual catastrophic loss of system, is surpassed. The latter leads to catastrophic fault. Then, using PBN, both classifications can be diagnosed and actions to mitigate them can be taken. The proposal also allows to forecast a time in hours by which the fault or failure will be imminent. The method herein discussed was applied to an ultrasound welding cycle, and a PBN with interventions model was created, simulated and verified through by means of model checking in PRISM. Results obtained show the validity of this methodology