Since the introduction of the industry 4.0 paradigm, manufacturing companies are investing in the development of algorithmic diagnostic solutions for their industrial equipment, relying on measured data and process models. However, process and fault models are not usually available for complex productions plants and production data are usually unlabeled. Thus, to classify machine status, unsupervised approaches such as anomaly detection and signal processing strategies have to be employed. Due to the unsupervised nature of the problem, it is meaningful to apply several diagnostic algorithms to cover most of the process anomalous behaviors. Additionally, in some contexts, the experience of process operators in grasping the correct functioning of machines as well as their ability in understanding early signs of deterioration is relevant for the diagnosis of incoming failures. However, seldom these information can be included in failure diagnosis algorithms. In this paper, we propose a diagnostic scheme for condition monitoring of mechanical components. The proposed scheme combines anomaly detection algorithms, envelope analysis of vibration data, and eventually additional qualitative information on machine functioning. The combination of all the fault indicators is obtained leveraging on a fuzzy inference system. The proposed scheme is experimentally validated on a steel making plant with real process data, making use of heuristic information such monitoring reports of machine health status.
File in questo prodotto:
Non ci sono file associati a questo prodotto.