The main contribution of this paper is the strategy to use motor drive system as multivariable sensor, together with multi data fusion method for health evaluation. In the diagnosis of mechatronics, external sensors cause problems in signal acquisition stage. To overcome these problems, signals from the control process of motor drive system are directly collected to achieve multivariable self-sensing. Moreover, data fusion methods are proposed to convert massive data into RGB channel false color image and position-based polar image to carry comprehensive information. The CNN model is used to classify various working conditions. Quantitative simulation and experiment on a ball screw driven by permanent magnet synchronous machines (PMSM) are conducted to validate the proposed method. The results show that images with comprehensive information perform better in the diagnosis of mechatronics.