Filter cloths are essential components in filter press machines, directly impacting process efficiency and operating costs in tailings treatment applications. In demanding mining environments they undergo rapid wear and require regular cleaning and replacement. Unexpected failures can lead to costly downtime and mechanical damage. Despite their importance, current monitoring practices must rely heavily on manual inspection, resulting in over-conservative and inefficient maintenance strategies, while still failing to reliably prevent unexpected failures.

This work presents an innovative inspection system developed by Diemme Filtration Srl in collaboration with University of Bologna, for the automated visual assessment of filter cloths. The system is designed as a modular linear scanner, capable of acquiring high-resolution images of the filter cloths. Its integration is minimally invasive, as it is conceived as an add-on to existing filter press models, and its operation does not affect standard cycle times. The imaging system is paired with an image processing pipeline, having at its core a machine learning-based anomaly detection algorithm trained to identify a broad range of cloth defects, from large tears to small scratches, enabling early fault detection without supervision. Overall, the system enables a new maintenance strategy based on targeted and timely intervention.

We present a case study...