A high precision torque sensor is used for extracting combustion timing information from cylinder individual pressure estimates constructed from the torque measurements. A combination of physics-based and data driven modeling is used where the physical part of the model is based on equations describing contributions of inertial and gas forces while the flexing of the crankshaft, which has rather complex dynamics, is modeled using the data driven approach. The first part of the study shows the derivation of the models and how well the torque at the sensor position can be estimated from the pressures in the four cylinders. The second part demonstrates how it is possible to reconstruct cylinder individual torque and pressure by inverting the pressure to torque model. Going from measured torque to pressure in each cylinder is not trivial since the inverted model is ill conditioned around top dead centre which causes large errors where the precision is the most needed. A parameterized combustion model is therefore introduced to improve the signal to noise ratio in the estimated parameters. The proposed method for detecting combustion demonstrated good results with a coefficient of determination of 0.95 against "true" combustion phasing.