Ability to have least failures in products on the field with minimum effort from the manufacturers is a major area of focus driven by Industry 4.0 initiatives. Amidst traditional methods of performing system/subsystem level tests often does not enable the complete coverage of a machine health performance predictions. This paper highlights a workable workflow that could be used as a template while considering system design especially employing Digital Twins that help in mimicking real-life scenarios early in the design cycle to increase product’s reliability as well as tend to near zero defects.
With currently available disruptive technologies, systems integrated multi-domain 'mechatronics' systems operating in closed-loop/close-interaction. This poses great challenge to system health monitoring as failure of any component can trigger catastrophic system failures. It may be the reason that component failures, as per some aerospace reports, are found to be major contributing factors to aircraft loss-of-control. Essentially, it is either too expensive or impossible to monitor every component or subsystem of a complex machine and the current state of the Integrated Health Monitoring Systems seem to be quite inadequate. In this paper, we propose an approach that combines the best of the diagnostics and feature extraction techniques coupled with Artificial Intelligence as a solution to address the challenges of Prognostics Health Management (PHM) for complex systems. The paper also documents a standard procedure to apply the right technologies/tools at every stage so that a clear process can be applied for any similar complex system across the product development life cycle.
In this paper we derive the health status of subsystems by looking at system level responses [1]. Distinguishing features are derived from the overall system level response through feature extraction methodologies and then fed into decision making frameworks that are implemented using both Convolutional Neural Networks [7, 18, 19], Machine Learning [4] and Deep Learning. Models are trained with distinguishable features through system simulations [20]. Employing rightly designed ML models provide the ability of classifying the failure modes as well as to analyze system faults/responses. Predictive modelling techniques are applied to the ML processed data to deliver useful prognostics on the criticality of the failure mode, RUL of the components/subsystems while system is in operation can be determined. The proposed concept can be easily adapted to various systems from varying domains [2]. The methodology evolved in this work can be easily extended for various use cases for instance in the Transportation domain the user can get alerts not only of failures ahead of time but also the remaining useful lifer as well as possible causes of such a failure. This would let prevent downtime of the overall vehicle/fleet and thereby ensures smooth operation of the entire service.
As a case study, the present work demonstrates the a DPHM solution applied to electrical energy generator where failure mode effects of subsystems and their effect on the overall system performance are studied using Modeling and Simulation techniques. The overall work would finally lead in demonstrating a working recommendation/advisory system that understand the behavior as if it was a pure Digital Twin [24] and thereby giving a quick turn around for different use cases like study/analysis/what if/predict behavior under various operating conditions with a high level of confidence before the changes are tried on a real system.