文章基本信息

标题：Stochastic model of forecasting spare parts demand
本地全文：下载
作者：Milojević Ivan S. ; Guberinić Rade V.
期刊名称：Vojnotehnicki glasnik / Military Technical Courier
印刷版ISSN：0042-8469
电子版ISSN：2217-4753
出版年度：2012
期号：6748
页码：216-234
语种：
出版社：Ministry of defence of the Republic of Serbia: University of defence in Belgrade
摘要：If demand is known for the whole planning period (complete information), then this type of demand or a supply system is deterministic. In the simplest cases, the demand per time unit is constant. If demand levels change over time following a precisely determined and pre-known principle, this type of demand is also classified as deterministic. This quality of demand is very rare. In most cases demand is the product of a process, for example TMS maintenance, whose progression cannot be predicted due to a number of factors influencing the process and causing random demand changes. In this case, a supply system must function according to the complete information and with a certain degree of uncertainty. In cases when demand may be defined by some of the laws of the probability theory, we are talking about stochastic demand and a stochastic supply system. Demand can be described by mathematical expectation, mathematical expectation and standard deviation, probability distribution or as a random process. However, there is usually a need for the most complex description, i.e. the complex random process because both intensity of demand and the probability distribution change during the observed intervals. The level of temporal (dynamic) series is traditionally considered as a complex phenomenon consisting of four components: - basic tendency of phenomenon development - cyclical impact (long-term, 'ancient') - seasonal effects - random fluctuations. The basic tendency of phenomenon development means a long-term evolution of phenomena. A function that expresses the trajectory of changes of the basic tendency of a phenomenon development in the form of the equation is called a trend. Often, the trend involves time regression; i.e. the coefficients of the proposed functions are often determined by the least squares method. To roughly determine the coefficients of the proposed function, the sum of three and three-point methods are also used. After checking the hypothesis of the existence of phenomenon change trends, the next step in the methodology of forecasting is the determination of a specific growth curve that describes the regularity of the development in time. These curves of growth are obtained by the analytical representation (expression) of dynamic lines. There are two basic stages in the process of expression and they are: - The choice of the type of curve the shape of which corresponds to the character of the dynamic order variation - the determination of the number of values (evaluation) of the curve parameters. The most widespread method of forecasting is the trend extrapolation. The basis of the trend extrapolation is the continuing of past trends in the future. The simplicity of the trend extrapolation process, on the one hand, and the absence of other information on the other hand, are the main reasons why the trend extrapolation is used for forecasting. The trend extrapolation is founded on the following assumptions: - The phenomenon development can be presented as an evolutionary trajectory or trend, - General conditions that influenced the trend development in the past will not undergo substantial changes in the future. Spare parts demand forecasting is constantly being done in all warehouses, workshops, and at all levels. Without demand forecasting, neither planning nor decision making can be done. Demand forecasting is the input for determining the level of reserve, size of the order, ordering cycles, etc. The question that arises is the one of the reliability and accuracy of a forecast and its effects. Forecasting 'by feeling' is not to be dismissed if there is nothing better, but in this case, one must be prepared for forecasting failures that cause unnecessary accumulation of certain spare parts, and also a chronic shortage of other spare parts. All this significantly increases costs and does not provide a satisfactory supply of spare parts. The main problem of the application of this model is that each spare part has its own failure intensity; therefore, it is necessary to conduct a complete procedure for each part. If the average number of components of a complex asset were about 5000, along with the existence of a large number of resources, the forecasting by the trend extrapolating would practically require the newest information technologies. However, the analysis shows that in most cases 80-90% of component parts have a longer life than the asset itself. In addition, it is shown that during the life of the asset, these components do not fail, and that 10-15% of component parts have only rare failures (post optimal analysis of spare parts normative) or do not show any failing tendency meaning that failures are random. This procedure remains to be applied to 2-4% of component parts. It does not present a significant problem due to the process automation. This model enables us to determine the parts to which the trend extrapolation procedure should be applied. Forecast is obtained by the use of the second model to existing data. The trend extrapolation forecast or any other kind of forecast does not need to be done for those spare parts for which the long time forecast is close to zero. Provision of spare parts for the maintenance system considering the range, quantity, time and place is crucial for successful functioning of the maintenance system and ultimately for the accuracy of material resources. On the other hand, stocks of spare parts are costly and required to be reduced. Costs need to be minimized while ensuring the successful functioning of the maintenance system.