Abstract: A electronic engine control (EEC) module executes a generic neural network processing program to perform one or more neural network control funtions. Each neural network funtion is defined by a unitary data structure which defines the network architecture, including the number of node layers, the number of nodes per layer, and the interconnections between nodes. In addition, the data structure holds weight values which determine the manner in which network signals are combined. The network definition data structures are created by a network training system which utilizes an external training processor which employs gradient methods to derive network weight values in accordance with a cost function which quantitatively defines system objectives and an identification network which is pretrained to provide gradient signals representative the behavior of the physical plant. The training processor executes training cycles asynchronously with the operation of the EEC module in a representative test vehicle.

Abstract: A method for identifying engine combustion failure of an internal combustion engine having a plurality of cylinders, a crankshaft and a crankshaft position sensor includes the steps of operating the internal combustion engine to rotate the crankshaft, measuring rotational quantities of the crankshaft corresponding to events created by each of the plurality of cylinders during operation of the internal combustion engine, correcting the rotational quantities measured to remove periodic position irregularities to generate a corrected temporal signal, generating an acceleration signal of the crankshaft using the corrected temporal signals, and identifying combustion failures as a function of the acceleration signal. A time-lagged recurrent neural network utilizes the acceleration signal, along with other engine parameters to identify the cylinder-specific misfire events.

Abstract: Irregularities in crankshaft velocity introduced when measuring crankshaft rotation at a section of a crankshaft in an internal combustion engine that is less damped to torsional oscillations than is another more accessible crankshaft section are corrected by performing a nonlinear transformation via a neural network to predict rotation measurements that would have been obtained at the inaccessible section from data actually collected at the accessible crankshaft section. Thus, the effects of torsional oscillations in the crankshaft are substantially filtered away, resulting in crankshaft acceleration values that form the basis of a misfire detector having nearly maximum signal-to-noise performance.