Abstract: A model-agnostic aggregation classifier meta-learning framework (MAACML) is proposed to address the issue of insufficient fault samples in current industrial processes for fault diagnosis methods based on multivariate statistical analysis and deep learning. Firstly, this framework combines model-agnostic meta-learning with convolutional neural networks and introduces an aggregation classifier to improve the classification accuracy and generalization ability of the model. Then, simulation experiments are conducted on the Tennessee Eastman simulation dataset to validate the performance of the model. Finally, the model is validated on an actual compressor unit dataset to evaluate its effectiveness. The research results demonstrate that the MAACML framework achieves higher average accuracy compared to other methods and exhibits good generalization ability. The introduced aggregation classifier module significantly enhances the classification results. The classification accuracy on the actual dataset reaches 100%, confirming the practicality and effectiveness of the MAACML framework.