Evaluation of the possibility of using artificial neural networks for self-diagnosis of an internal combustion engine with cylinder deactivation

Abstract

BACKGROUND: Engine and vehicle control systems must have means of objective control in the form of self-diagnosis. This is especially true for new designs and technologies for controlling a gasoline internal combustion engine, such as deactivating cylinders in partial load mode. The paper gives an assessment of the possibility of self-diagnosis of cylinder shutdown in an automobile crank-guide engine without connecting rods using artificial neural networks.
 AIMS: Determination of the possibility of creating an artificial neural network that recognizes which cylinders are currently in operation and which are disabled, based on the nature of the change in the signals from the sensors installed on the engine mounts and independent on the crankshaft speed.
 METHODS: The study considered artificial neural networks of the LSTM and BiLSTM topology. An engine simulation model made in Simulink was used in order to obtain sensor signals. The conducted numerical experiments made it possible to obtain data, which simulates the sensors readings, and to train artificial neural networks to determine the order numbers and quantity of deactivated cylinders. Numerical experiments were carried out on the basis of full-factorial design. Various designs of experiments were used for training and testing of artificial neural networks, which made it possible to test the network on data that differed from the training data significantly. Testing took place on a large number of random sequences of cylinder deactivation modes.
 RESULTS: The obtained results show a high degree of recognition of the order numbers of deactivated cylinders just after several tens of degrees of the crankshaft rotation while switching to the corresponding mode. For the LSTM network, mode detection accuracy was above 99% in both the data sequence transfer mode and the data streaming mode. Accuracy of the BiLSTM topology was over 99.9% in the data sequence transfer mode, but significantly decreased in the data streaming mode.
 CONCLUSIONS: The use of considered types of networks in engine and car control systems is promising.