Simulation et analyse parametrique de methodes de prise de decision dans le cadre de la maintenance conditionnelle

With the increasing dependence on automation and the massive investment in physical assets, the organization are looking for more accurate, more cost effective maintenance management strategies. The run-to-failure strategy is no longer acceptable. The preventive and the predictive maintenance strategies are none parts of physical assets management strategies in many companies. An important part of a predictive maintenance strategy is the condition monitoring and the condition based maintenance. These facts lead the researcher to propose different decision support models. Probabilistic models can help the decision makers in taking the right decisions at the right time. But as any probabilistic statistically dependant decision modes, they are inherently subjected to two type of error, the false negative, which leads to letting the system work as it is while it is supposed to be maintained and the false positive, which leads to interrupting the system while it is working properly. Using any statistically based decision making models without taking into consideration these two types of errors may lead to erroneous decisions.; The purpose of this thesis is to calculate and analyze the false negative and false positive errors that are inherent to the probabilistically and statistically dependent decision making models used in condition based maintenance management. Specifically, three models are analyzed: the Statistical Process Control (SPC), the Hidden Markov Model (HMM) and the Proportional Hazards Model (PHM). These three models appear to be the most presented in the literature. This thesis presents a study of the change in the two types of error, the false positive and the false negative, when the three models parameters change.; The methodology used is based on a computer simulation approach integrating the SPC, the HMM and the PHM methods. The simulated models are run for a predefined number of times, while the models parameters are held at a predefined level. A specific design of experiments based on Taguchi's method allowed us to limit the number of runs considerably. The results obtained are the analyzed by using the Analysis of Variance (ANOVA) technique.; To build the simulations models, some assumptions were necessary. Without loss of generality, it was assumed that the system can be in only one of two states, a working state or a deteriorating state, which cannot be detected visually. It was also assumed that the observations collected from the system are random variables that follow normal distribution with means and variances that vary according to the states. It was also assumed that the probability of changing from on state to the other in the period between two consecutive observations is constant, and that the change of state modifies only the mean value of the normal distributions, the variance remains unchanged.; The results show that under the simulated conditions, the SPC method has the lowest percentage of the two types of errors, even if the system is in the deteriorating state is not valid. The HMM methods provides acceptable level for each type of error separately but not simultaneously. Its performance is quite sensitive to the difference between the process means before and after deterioration. Finally, the PHM performance was similar to the HMM method. The decrease in the false positive error leads to the increase in false negative error, and the difference between the process means before and after the deterioration has an important effect on the accuracy of the decisions made.; This work showed that when using any of the three methods of decision making in condition based maintenance (CBM), the SPC, the HMM and the PHM, the user should calculate the false positive and the false negative rates that are inherent to these methods. The research shows that under certain condition and for some values of the models parameters, these error increase. The effect of these errors on the financial aspects of deci...