Abstract The detection and isolation of faults in engineering systems has lately become of great significance. This paper is concerned with application of analytical fault detection techniques to a heat exchanger. The system is nonlinear and a velocity-based linearization is proposed before the residual generation, which is then realized using an observer, and parity relations. The problem of reasoning is treated by an approximate reasoning approach, called the transferable belief model. Its important feature is the ability to treat inconsistency in data by using more general belief functions. The system under consideration is controlled by programmable logic controller and supervised by supervisory control and data acquisition system. The procedure for implementation is designed with particular emphasis on industrial practice and the Ole for process control data access client/server technology is used for communication between sub-systems.