Comprehensive anomaly detection method for predictive maintenance of belt conveyors

Belt conveyors are critical transport systems in the mining, manufacturing, and logistics industries. Unforeseen failures lead to costly downtime and safety risks. Existing monitoring systems based on traditional threshold methods often trigger false alarms due to transient process disturbances and fail to provide timely anomaly detection. In this regard, the development of signal filtering and analysis methods robust to transient noise is a pressing task. The aim of the study is to develop and substantiate a comprehensive automated monitoring method based on a multisensor system and temporal data processing algorithms for early detection of anomalies and the prevention of emergency situations. The proposed method integrates a multi-sensor system comprising tension, misalignment, overload, sag, and acoustic sensors with a centralized control unit. Data processing is performed in the time domain using adaptive threshold values derived from the statistical characteristics of signals obtained during normal conveyor operation. Anomaly detection is based on analyzing consecutive threshold exceedances within a specified time window, which mitigates the impact of random noise and transient spikes. An architecture for a multisensor monitoring system featuring redundancy for critical sensors has been proposed. A time-domain data processing method utilizing adaptive thresholds has been developed to filter out random noise and sporadic spikes. It has been theoretically and experimentally demonstrated that the decision-making procedure based on a series of consecutive exceedances within a specified time window enables the separation of anomalous events from random noise. It is demonstrated that employing a criterion based on the accumulation of consecutive exceedances ensures more robust detection of belt misalignment, overloads, and anomalous acoustic signals compared to monitoring individual threshold violations. The method enables real-time data recording and supports predictive analytics through the integration of machine learning algorithms. The results achieved fulfill the research objective: to enhance belt conveyor reliability and establish a foundation for condition-based maintenance strategies. The practical value lies in reducing unplanned downtime and mitigating safety risks at industrial facilities. Future work will focus on experimentally evaluating the proposed method at operational production sites, conducting a cost-benefit analysis of its implementation, and subsequently integrating machine learning techniques to advance predictive analytics capabilities.