Unsupervised machine learning model for resistance variant fault diagnosis of mine ventilation system with composite features

The present resistance variant fault diagnosis method for mine ventilation system needs to collect fault samples before the fault location and fault volume diagnosis. Furthermore,the mathematical model for classification and regression should be built for the location and volume of the fault diagnosis separately. In terms of the issues of the resistance variant fault diagnosis method,such as,difficulties in collecting the fault samples,and diagnosing the fault location and volume at the same time,the resistance variant fault diagnosis of mine ventilation system was transformed into an optimal problem of minimum Euclidean distance,thus an unsupervised learning model for resistance variant fault diagnosis of mine ventilation system without the training sample was proposed,and the Covariance Matrix Adaptation Evolution Strategy method was used to optimize the unsupervised learning model to realize the integration of classification and regression prediction. A comparable simulation test of air volume,wind pressure single feature and air volume-wind pressure composite characteristics was conducted and the results show that the proposed fault diagnosis model and the solution method can effectively solve the fault diagnosis problem of mine ventilation system resistive type without sample participation. In the process of fault diagnosis,it is not necessary to diagnose fault location and the fault magnitude separately. Compared with the resistance variant fault diagnosis of mine ventilation system under the single feature of air volume or wind pressure,the composite features of air volume-wind pressure can achieve a higher accuracy of fault location diagnosis and a lower error of fault volume diagnosis. Even if some observation points are selected,the model can achieve a higher fault location diagnostic accuracy and a lower fault diagnostic error,and the proportion of fault observation points has no direct influence on the diagnostic effect.