A crank press is converted into a smart fatigue testing machine for 3-point bending of polymer matrix composite specimens. The press is retrofitted with a load cell base for work holding, which monitors the bending force applied by the ram, a cycle counter recording the number of loading cycles, and a camera recording snapshots of the specimen area where failure is expected. A convolutional and Resnet neural networks are trained to recognize failure as an area of color change in camera images. Signal drop on the load cell signals failure onset, thus triggering monitoring by the camera and execution of the neural network. Acceptable proof-of-concept results encourage further automation of the setup.