Intelligent System Monitoring: On-Line Learning and System Condition State

Intelligent System Monitoring: On-Line Learning and System Condition State

Claudia Maria García (Universitat Politècnica de Catalunya UPC, Spain)
DOI: 10.4018/978-1-4666-2095-7.ch002
OnDemand PDF Download:


A general methodology for intelligent system monitoring is proposed in this chapter. The methodology combines degradation hybrid automata to system degradation tracking and a nonlinear adaptive model for model-based diagnosis and prognosis purposes. The principal idea behind this approach is monitoring the plant for any off-nominal system behavior due a wear or degradation. The system degradation is divided in subspaces, from fully functional, nominal, or faultless mode to no functionality mode, failure. The degradation hybrid automata, uses a nonlinear adaptive model for continuous flow dynamics and a system condition guard to transition between modes. Error Filtering On- line Learning (EFOL) scheme is introduced to design a parametric model and adaptive low in such a way that the unknown part of the adaptive model function is on-line approximated; the on-line approximation is via a Radial Basis Function Neural Network (RBFNN). To validate the proposed methodology, a complete conveyor belt simulator, based on a real system, is designed on Simulink; the degradation is characterized using the Paris-Erdogan crack growth function. Once the simulator is designed the measured current, i’s, and velocity of the IM, ?m, are used to modeling the simplified adaptive IM model. EFOL scheme is used to on-line approximate the unknown TL function. The simplified adaptive model estimates the IM velocity, , as output. and the measured IM velocity ?m are compared to detect any deviation from the nominal system behavior. When the degradation automata detect a system condition change the adaptive model on-line approximate the new TL.
Chapter Preview


The Prognosis and Health Management (PHM) field is predicated on four fundamental notions (Uckun, Goebel, & Lucas, 2008): first, all electromechanical systems ages as a function of use, passage of time and environmental conditions; second, component aging and damage accumulation is a monotonic process, and it’s shown at the physical or chemical composition of the component; third, signs of aging are detectable prior to over fatal or total failure; and finally, it is possible to correlate ageing signs with a component or system model.

Based on this four predicates, Figure 1 shows a proposed intelligent monitoring system bloc diagram. The diagnosis and prognosis is based on an adaptive system model which is on-line approximated to follow the system degradation. The system monitoring is due by hybrid automata, which change mode depending on the system condition stage:

Figure 1.

Intelligent monitoring system bloc diagram

  • Step 1: Data acquisition: this procedure is used to obtain the different signals available in the system. It covers the range of normal operation to fault and degradation behavior of the plant.

  • Step 2: Simplified adaptive model: From mathematical viewpoint the selection of a function approximator provides a way for parametrizing unknown parts of the physical model; designing on-line learning algorithms and parameter adaptive laws.

  • Step 3: Diagnosis: Allows comparing the acquired data and the physic model to detect any off-nominal system situation.

  • Step 4: System monitoring: It is based in the proposed degradation hybrid automata; allowing system condition tracking.

  • Step 5: Prognosis: based on actual data and the physic model, prognosis techniques are capable of forecast the future system behavior.

The main area of interest in this chapter is the impact and potential benefit of use nonlinear parametric models to approximate unknown functions in such a way that the descriptive physical model can be on-line adaptable. The use of on-line learning techniques allows the use of the degradation hybrid automata for system condition monitoring. The unknown function approximation is done by using a Radial Basis Function Neural Network (RBFNN). Diagnosis and Prognosis are out of scope, centering this work in the degradation automata for intelligent monitoring, step 4, and parametric estimator to adapt the model under degradation, step 2.

The methodology is exemplified via a complete conveyor belt system designed in Simulink and based on a real system. The Simulink conveyor belt process is employ to carry out fatigue or wear simulations; the process degradation is introduced by two friction parameters, , the static and kinetic degradation friction respectively, which are time varying and follow the Paris-Erdogan crack growth function. The conveyor belt system is split in sub-parts: a nonlinear nominal Induction Motor (IM), a scalar and vector control law, a thermal model and the conveyor belt physic description. The monitoring methodology is design a simplified adaptive model, for the electric and mechanic part of the IM, using the IM velocity and current as inputs and estimating the motor velocity as output.

Complete Chapter List

Search this Book: