Intelligent computing methods for indicated torque reconstruction

The paper proposes using a support vector machine to reconstruct indicated torque from the crank angle signal in an automotive engine. Support vector machines have been shown to perform extremely well in many classification and regression applications. The relationship between indicated torque and crankshaft angular velocity is a current research topic, and is also a nonlinear problem. In a typical combustion engine, cycle-by-cycle variations of combustion events occur, even when running at a fixed operating point. Engine idle speed controllers capable of reducing the variability have been proposed, and rely on indicated torque information. Furthermore, real-time indicated torque knowledge is important for engine diagnostics. The proposed approach provides the potential for real-time reconstruction of indicated torque and reduction in costs for manufacturers. A comparison between the proposed approach and another popular model estimation approach, K-means clustering with RBF centres trained using a least mean squares algorithm, is presented. Reduction in the input data resolution and its effect on reconstruction accuracy is also investigated.