Health diagnosis for rotating machines significantly contributes to improving safety and reliability. This research studies an AIOT system for anomaly prognosis for diesel generators utilizing deep learning algorithms, the convolution neural network, and bidirectional long short-term memory (CNN-BiLSTM) algorithms. Firstly, the IoT systems are designed to collect the simulated anomaly condition data of industrial 125 kW/250 kW diesel generators in the laboratory, which are utilized as training label data. The CNN combined BiLSTM captures the mapping correlation between selected features through the training operations to prognosis the anomaly conditions of the industrial diesel generator. The superiority of the developed algorithm is verified with other deep learning approaches, including the recurrent neural network (RNN), the gate recurrent unit (GRU), the LSTM, and the CNN. An experiment verification is deployed to evaluate the effectiveness of the hybrid methodology, which proves the excellent ability and demonstrates the superiority in anomaly condition classification compared with other individual deep learning models. Through different experiment datasets, the proposed methodology’s improved performances demonstrated its contribution to a novel framework beyond other advanced algorithms. The application of the developed AIot-based CNN-BiLSTM algorithm has been successfully deployed for anomaly conditional classification in the industrial 125 kW/250 kW diesel generators in Taiwan.