Algorithm for Reliability Centered Maintenance Modeling and Analysis
- Step 1: Identification and Definition of Problem or Setting the Hypothesis
Usually, the reliability studies consume much cost and time. In addition, it requires higher level engineering knowledge to understand the complex problems and models. Therefore, it should be first decided whether the study is worthwhile. Hence it is often suggested to select the area where the cost of breakdown is very high or any failure is hazardous or leads to catastrophe that may claim human lives, assets and/or environmental disorders. Once the decision is taken, problem should be defined clearly and the bounds of the study is also be demarcated, otherwise it becomes endless project. However, improvements can always be incorporated at the later stages based on the feedback reports. A suitable hypothesis is stated for the equipment upon which the reliability studies are to be conducted.
- Step 2: Collection of Relevant Data
Relevant data is collected often in the form of Time Between Failures, in short called TBF and Time To Repair, concisely called TTR (in the case Repairable items) or Time To Fail, in short called TTF (in case of non-repairable items). While collecting such field data there may exists some practical problems. Suitable assumptions as applicable may be made to overcome such problems. The data so collected is to be tabulated.
- Step 3: Elusion of Data Inconsistencies and Errors
To err is human. There is every possibility of occurrence of errors in the collected data. Such suspicious entries may be eluded. Further, in some situations there may be some inconsistencies in the data of TBF values. For instance, a particular machine (say an automobile) may be under breakdown/non-available for the want of some spare parts (say a valve for a tyre). If there is any delay in supply of such spare part, the machine can be restored to up condition and the downtime will increase. Suppose, this continues for a very long period, the machine should be deleted from the list completely. Similar inconsistencies, if found should be precluded for the analysis.
- Step 4: Trend Analysis and Correlation Tests
These are already discussed in the previous units. The trend analysis gives us the idea about whether the machine is deteriorating or improving, by which we can also understand which phase of life cycle (Bath tub curve), the equipment is experiencing (Infant/youth/old age). Various graphical (Eye-Ball analysis, Cumulative plot test, etc.) and analytical methods (Laplace Test, Mil-hdbk-189 test) are available for testing the trend. The correlation and regression analyses also can aid the study in confirming the time independence of the distribution. If the trend is positive, assumption of independently and identically distribution (IID) is said to be not contradicted. Hence, we can fit the data in Non-Homogeneous Poisson Process (NHPP) models such as Power Law Process (PLP). In contrary, if the data does not exhibit any trend (or very weak trend which is not considerable) then we have to go for further analysis by Total Time on Test (TTT) potting or other suitable reliability models.
- Step 5: TTT Plots to Examine Exponential Fit
Total Time on Test (TTT) graphs can be plotted in several ways. The most popular and convenient methods are Maximum Likelihood Estimation (MLE), Piecewise Exponential Estimation (PEXE) and Product Limit Estimation (PLE) or Kaplan-Meyer Estimation (KME). With the help of TTT plots, we will be able to know whether the data has any exponential fit. If there is exponential fit, we can model the data in Homogeneous Poisson Process (HPP), otherwise it can be fit in Renewable Process (RP) such as Weibull Model.
- Step 6: Fitting the Suitable Model
Using relevant statistical tools, the data is fit in suitable model. Its characteristics can be hence studied to design the reliability of the system.
- Step 7: Confirmation Tests and Goodness of Fit
Further, the fit of the model can be confirmed by various statistical methods (Hypothesis tests). The general Goodness of fit tests such as Chi-square tests, are universally applicable while some specific tests are available to check particular distributions. The specific goodness of fit tests such as Bartlet's test (Exponential Distribution), Mann's test (for Weibull Distribution) and Kolmogorov-Smirnov test (for Normal/Lognormal Distribution) are the few found among most useful tools in this study. The goodness of test not only confirms the distribution but can authenticate the methodology of reliability analysis.
- Step 8: Reliability Centered Maintenance Planning and Scheduling
The reliability characteristics are then estimated. Various characteristics such as Toptimal, Tmedian, Tmode, B1 life, B. 1 life, etc. can be estimated and hence maintenance planning and scheduling can be done.
- Step 9: Reliability Growth Testing
Further the Reliability Growth can be tested using suitable growth models such as Duane's growth model, AMSAA models in test-fix-test-fix cycle.