Probabilistic assessment of availability from system performance data

摘要

The traditional means of availability assessment as a probability distribution are not widely used on manufacturing plants and facilities producing goods and services, in part because of a lack of accurate data. An alternative method was developed for such plants using permutations of readily available system data. The method may be found to be of value for other repairable systems. This paper presents a method of calculating system availability and reliability probability distributions using permutations of inseparable system failure and restore data sets. Such data sets usually come from system history, as reflected in a performance measurement such as daily production. A direct relationship is maintained between any failure and its consequence; that is, time-between-failure (TBF) and time-to-restore (TTR) are an in set. Furthermore, TTR need not be single-valued but may take the form of another data set to accommodate system degradation and dependent failures. A time line can be developed (in a computer spreadsheet or as a mathematical concept) on which the data sets are placed. A time window (W) equal to a time interval of interest (mission time) is advanced along the time line returning an availability discrete random variable value at each position. In general, there are {H(N!) -(W-1)} mission times of length W in history of length H containing N independent failures. For example, 12 failures in two years provide about 350 billion mission time values. From these values, or a sample of the values, availability frequency distributions are formed for all mission times of interest. When the mission times of interest are continuous, as is necessary for certain business decisions, the distributions form a 3-dimensional probability surface. From this data, both the probability and expected magnitude of performance below or above any value is calculated for use in making a large range of technical and business decisions. The avoidance of traditional assumptions, the accuracy of calculation and the abundant supply of accounting quality data provide an opportunity to make risk-based decisions that are not otherwise possible. For example, in manufacturing the issues which can now be optimized with known probability and consequence include production budgeting, product inventory control, profit projections, material requirements planning, production scheduling and measuring the statistical significance of any change in production output. This capacity and availability assessment process (CAAP) is patented in the United States for systems producing products and services, such as manufacturing, telecommunications, power generation and other utilities.