<p>The firefighting pumps have been extensively utilized in commercial, residential, and industrial buildings due to their critical role in supplying high-pressure water for fire protection functions. Ensuring the availability and reliability of these pumps during emergency situations is a vital task for maintenance engineers, who typically rely on periodic inspections. This study proposes an Artificial Intelligence of Things (AIoT)-based fault diagnosis framework for firefighting pump maintenance using a hybrid convolutional neural network–long short-term memory (CNN-LSTM) algorithm. The proposed framework performs real-time monitoring and early fault identification through continuous data acquisition from IoT modules equipped with multiple high-precision sensors installed on firefighting pumps under various operational and failure conditions. The CNN-LSTM model is designed for adaptive online learning, enabling it to process real-time data streams and support scalable deployment across multiple pump systems. Furthermore, the framework is compatible with existing maintenance protocols by providing interpretable diagnostic results for timely decision-making. The model’s performance is validated against traditional algorithms, including recurrent neural network (RNN), CNN, gate recurrent unit, LSTM, and CNN-RNN methods, using datasets collected at 10 s, 30 s, and 1 min intervals. Experimental results demonstrate that the proposed AIoT-based hybrid CNN-LSTM approach achieves superior accuracy and robustness, offering an intelligent and scalable solution to reduce manual inspection efforts and enhance maintenance efficiency. </p>