Revisiting a Predictive Maintenance Toolbox: Sensor Retrofit and Workflow Demonstration with a Paired Sensor Study

Inhalt:
Predictive maintenance (PdM) adoption is often limited by sensor cost, integration effort, and the opacity of proprietary solutions that are hard to reproduce without implementation details. We present a paired, within-process evaluation of a low-cost MEMS accelerometer versus a higher-cost, higher-spec accelerometer for tool-state classification in milling. Both sensors are co-located and synchronized to record identical cuts across two tool conditions (good and bad). We follow a transparent, step-by-step PdM workflow: retrofit and data capture, band-limited feature extraction, and a simple, well-documented machine learning (ML) pipeline with best-practice tips. The ML pipeline is applied identically to both datasets and fixed before testing on a separate dataset collected on a different day. This puts the two sensors head-to-head in a direct comparison. The results show how an open, reproducible workflow can make performance-vs-cost decisions explicit and actionable, highlighting when low-cost sensing is sufficient and when higher-end hardware is warranted. In our case, the low-cost sensing proved sufficient for the task.