Multisine Input Design for Active Data-Driven Fault Diagnosis of Belt Drives

Abstract

Belt drives have versatile industrial applications. A proper pretension is necessary to achieve high efficiency and low wear. For this purpose, active fault diagnosis (FD), where an auxiliary signal is injected, has shown promising results. But published approaches for input design require high domain knowledge, making them impractical in many real-world applications. We propose a procedure for input design in a data-driven FD setup and apply it to a real-world application. Multisine signals are optimized to achieve maximum separability showing significant performance improvement compared to passive FD. The resulting input signal leads to a high system disturbance which is undesirable if injected during normal operation. A minimum energy signal that still ensures successful FD is designed to solve this problem. In this way, AFD systems are superior to passive approaches while minimizing their downside of disturbing the machine operation. As a result, AFD's feasibility and potential are proven leading to increased reliability of belt drives.