Aiming at the problem of frequent aging faults of insulated gate bipolar transistor modules, this paper takes IGBT aging fault characteristic parameter data collected from the intelligent sensing layer of the Internet of Things terminal as samples. In the computing fusion cooperation layer, multi-terminal cooperation training and establishment of an optimized random forest model is applied to the IGBT aging fault diagnosis system to realize the condition monitoring of IGBT devices. Firstly, the characteristic parameters of aging fault are selected from the data samples and preprocessed to establish the aging fault diagnosis data set. Secondly, the traditional random forest model is built and optimized by parameter optimization of base evaluator, parameter optimization of model frame and bagging method. The optimization model training was completed on the basis of cross-validation. Finally, the prediction effect of the model in this paper and other models on IGBT aging fault diagnosis data set was evaluated by various evaluation indexes. Finally, the optimized model fit well, the error between the training set curve and the test set curve was 1.19%, and the prediction accuracy on the test set could reach 98.81%. The feasibility and accuracy of the optimized random forest model applied to IGBT condition monitoring system in the Internet of Things environment are verified.