Forecasting spare parts demand using condition monitoring information

Abstract

The control of an inventory where spare parts demand is infrequent has always been complex to manage because of the randomness of the demand, as well as the existence of a large proportion of zero values in the demand pattern. However, considering the importance of spare parts demand forecasting in production manufacturing and inventory management, several forecasting methods have been developed over the years to allow decision makers in industry to optimize the management of inventory where the demand pattern is infrequent. The Croston method is one of the traditional forecasting method, known because of its ability to take into consideration periods with zero demands. Yet, despite the Croston method’s advantage over other traditional methods, there are still shortcomings in the method because it does not consider the condition of the components to be replaced. This dissertation proposes an alternative forecasting method to the traditional methods, by means of condition monitoring. This method overcomes the Croston method’s shortcomings by considering the condition information of the component under operation. A statistical model, the so-called proportional hazards model (PHM), which is a regression model, blending event and condition monitoring data, is used to estimate the risk of failure for the component under analysis, while subjected to condition monitoring. To obtain optimal decision making on spare parts demand, a blending of the hazard or risk with the economics is performed, and an optimal risk point is specified. The optimal risk point guides optimal decision making on spare parts policy for the component under analysis. To generate the data needed to construct the proportional hazards model, a numerical investigation was performed on a fan axial bade where a crack was inserted and propagated to estimate the fatigue crack life and corresponding natural frequencies. The simulation was run using MSC.MARC/MENTAT 2016 software. To validate the finite element model, an experiment was run by using a 50kN Spectral Dynamics electrodynamics shaker to apply base excitation to the fan axial blade specimens. The treatment and computation of data generated from experimental and numerical approaches allowed the construction of the proportional hazards model, with the fatigue lifetime as event data and the blade natural frequencies as covariates or condition monitoring information. The baseline Weibull parameters were estimated by maximizing the likelihood function using the Newton Raphson method and the MATLAB package. This allowed the computation of an objective function to determine the shape, scale and location parameters. Instead of defining the covariate behaviour needed to build the cost function by means of the Markov process, a simulation procedure was utilized to define the cost function and determine the optimal risk which minimizes the cost. Furthermore, as the proportional hazards model depends on both, time and covariates, it was also shown how the PHM behaves when time or covariates carry more weight. The added value of the proportional hazard model as forecasting spare parts method lies in the fact that it allows one to proactively gather failure information which enables a ‘just in time’ supply of spare parts as well as an optimal maintenance plan. Forecasting spare parts demand, using condition information, performs better than traditional methods because it reduces an overly large spare parts stock pile. By allowing a ‘just in time’ part availability, the spare parts management becomes more related to the condition of the asset. Additionally, the supply chain management and maintenance cost are optimized, and the preventive replacement of components is optimized compared to the time-based method where a component can be replaced while still having a useful life.