PREDICTION OF VIBRATION ANOMALIES IN INDUSTRIAL EQUIPMENT USING COMPUTER VISION AND CUBIC SPLINE INTERPOLATION

Abstract

This article presents a hybrid approach for early prediction of vibration anomalies in industrial equipment using ordinary video frames and numerical signal reconstruction methods. Contact sensors, including accelerometers, provide accurate measurements, but their installation, calibration, and maintenance across many objects are costly. The proposed model first estimates inter-frame displacement through optical flow and then smooths the noisy, irregular low-FPS signal with cubic spline interpolation. Amplitude growth over a longer interval is approximated with the least squares method, and remaining useful life (RUL) is then estimated from the fitted trend. The experimental part is not based on long-term measurements from a real industrial object; it uses simulated MQTT mock streams aligned with the SensorFusionArbiter and MqttIngestService structures. The results show that global RMSE decreased from 0.118 mm for the raw optical-flow signal to 0.042 mm after spline processing, while MSE reached 0.0017 mm². The C++ native engine processed one frame in an average range of 12–14 ms on an Intel Core i9-14900KF and NVIDIA RTX 5080 configuration.  Laboratory and field validation under real industrial conditions is considered the next stage of the study