This paper focuses on gearbox wear condition monitoring, addressing issues with traditional techniques. We built a multi-source heterogeneous information online monitoring big data platform for comprehensive gearbox wear status monitoring. A Gaussian mixture model-based abrasive segmentation algorithm and wavelet thresholding method were proposed for feature extraction. A KPCA-PSO-LSTM evaluation method was adopted, with experimental results showing higher accuracy for mixed feature data. A web application for real-time monitoring was also developed. However, there are still shortcomings, and future work includes continuous model training using big data, efficient information integration, and feature expansion. This study provides new ideas and technical support for gearbox wear monitoring, contributing to equipment health management.