Dynamic logistic regression and electromechanical coupling analysis for remanufacturing suitability assessment of retired motors

In response to the problems of incomplete quality evaluation system and low detection efficiency in the remanufacturing process of retired motors, this study proposes a quantitative evaluation method for the applicability of retired motors based on dynamic coupling analysis of electromechanical parameters. By collecting mechanical parameters (stator-rotor deflection angle θ) and electrical parameters (three-phase inductance difference rate ΔL , impedance difference rate ΔZ , harmonic difference rate ΔI / F , and ground insulation resistance Rg ) of 47 retired drive motors for new energy vehicles, a multidimensional electromechanical collaborative degradation characteristic system was constructed, and a logistic regression model with dynamic weight adjustment was designed. This method innovatively introduces a condition adaptive dynamic weight mechanism: when θ>0.2 rad, the mechanical weight gain is 40%; When the insulation resistance Rg is less than 5 MΩ, the electrical weight sensitivity is increased by 35%, effectively solving the limitations of independent analysis of electromechanical parameters in traditional methods; Extract key features reflecting the degradation of winding symmetry and conductor oxidation by removing extremely degraded samples with Rg <0.1 MΩ through data cleaning and Min-Max normalization; Combining L2 regularization with maximum likelihood estimation to avoid the risk of overfitting in small samples. The experiment showed that the accuracy of the model on the test set reached 92.3%, which was better than support vector machine (85.1%), random forest (88.2%), and gradient boosting tree (89.5%). The 5-fold cross validation showed excellent stability (accuracy 90.5% ± 1.8%, AUC 0.921 ± 0.014), and the misjudgment rate of boundary samples (60<score<80) was reduced by 23.5%; In industrial validation, the evaluation time has been reduced from 63.5 seconds for manual detection to 6.2 seconds, and the misjudgment rate of typical cases has decreased by 18%. This study provides a high-precision and low-cost standardized solution for the rapid classification and resource utilization of retired motors, which has significant engineering application value.