NETWORK-INDUCED DELAY MODELS FOR CAN-BASED NETWORKED CONTROL SYSTEMS

Networked Control Systems (NCS) are a variation of traditional point-to-point control systems where sensors and actuators may be physically distributed and a serial common-bus communication network is used to exchange system information and control signals. Because all components use the same communication network, network-induced delays could make the system stochastic. The Quality of Control (QoC) of each closed-loop system in a NCS is strongly affected by the network-induced delay produced by sensors and control signals. Controller Area Network (CAN) is a popular real-time fieldbus used for small-scale distributed environments such as automobiles. In CAN the delay exhibits a stochastic behavior and varies according to the network load. Since QoC is affected by delays, designing and evaluating a controller must take into account the effect of network-induced delays. This paper illustrates three models that play the role of classifiers and estimators of the network-induced delays; the models can estimate the network load and predict future time delay values. The models were built following a statistical approach using a continuous Hidden Markov Model, a black-box state space modeling approach using Recursive Multi-Layer Perceptrons, and a histrogram-based approach. Each approach was trained/tested using experimental data taken from a real CAN system with excellent results.