Method for recognizing abnormal operation patterns in hydraulic support machine-following and shifting control

In intelligent mining faces, abnormal operating conditions such as support loss and poor linearity may occur during automatic follow-up of hydraulic supports, due to various factors including the floor conditions, hydraulic system, and electro-hydraulic control system. Currently, the identification and analysis of these abnormal conditions mainly occur after the automatic follow-up process, relying solely on manual adjustments for judgment. This approach is inefficient for quickly determining which hydraulic support units require manual intervention. To address this issue, a method was proposed for identifying abnormal condition patterns in machine-following and shifting control of hydraulic supports. This method shifted the scope of anomaly detection to the shifting stage after the support has lowered its columns, and classifies abnormal patterns into three types: shutdown fluctuation type, over-time shifting type, and stroke anomaly type. Data on hydraulic support cylinder stroke and leg pressure were collected. Then, data preprocessing was performed, including outlier removal, calculation of differences between adjacent data points, and merging data based on the sign of the differences. An anomaly recognition algorithm based on stroke-pressure-time analysis was used to identify the shutdown fluctuation and over-time shifting types. A decision tree model was employed to detect stroke anomaly patterns. Six key features were extracted: start and end time of the shifting action, maximum and minimum values of the current support's stroke, and the stroke differences between the current support and its neighboring supports. The start and end times were fed into the stroke anomaly recognition algorithm to detect fluctuation patterns. For data with stroke fluctuations, further analysis of pressure variation and shifting action duration was conducted to identify shutdown fluctuation and over-time shifting types. The remaining four stroke-related features were input into the decision tree model to identify stroke anomaly patterns. Experimental results show that the proposed method achieved a precision of 100% and recall rate of over 95% for identifying shutdown fluctuation and over-time shifting types. For the sudden drop type stroke anomaly, the method achieved a precision of 100% and a recall rate of 97.87%. For the uniform small stroke anomaly, the precision was 95.29% and the recall rate was 81%. These results demonstrate that the method effectively identifies abnormal conditions in hydraulic support machine-following and shifting control.