US Patent 11053880 Abnormality diagnosis system for air flow meter

A storage device of a control system stores a trained neural network that is trained by using training data to which information as to whether there is an abnormality in an air flow meter is given as a true label. The trained neural network receives, as inputs, condition index values including the opening of a throttle valve and the engine speed, a load detected value calculated based on a detection signal of the air flow meter, a load estimated value estimated without using the detection signal of the air flow meter, and a change index value indicating the degree of change of the engine load. A CPU of the control system executes a diagnostic process to determine whether there is an abnormality in the air flow meter, using the neural network, which receives the condition index values, load detected value, load estimated value, and change index value as the inputs.