Condition monitoring for machine health prognosis using dominance based rough sets

  • Faith Thompson

Student thesis: Doctoral Thesis


Within the dairy industry there is no unique maintenance policy designed to handle all the tasks or situations, so a reactive maintenance of “the right strategy in the right situation” policy has been adopted. This thesis provides an online, automated software platform capable of assessing machine health to facilitate a change from the current reactive maintenance policy to a condition based maintenance policy.
Multiple different decision making methods were considered for the system, such as neural networks, expert systems and fuzzy systems and were discounted. Several advantages of Dominance-based Rough Set Approach (DRSA) over these methods made it an obvious choice for the decision-making technique embedded in the condition monitoring system. For example, the output of DRSA takes the form of logic statements or rules, which need no interpretation from experts or specialists, they are simple to implement in terms of computation complexity, and they can address hesitancy, ambiguity and vagueness in the data and in the preferences of the classes by the distinction of different kinds of decision rules.
Implementing DRSA in a three phase, multi-criteria, iterative, classification framework, has been proposed. During the first phase historical, live industrial data is used as a learning set for the DRSA, and a set of conditional statements are generated to classify data. The second phase validates the conditional statements generated, using a combination of automatic and manual techniques. Finally the third stage of the process is to classify, current, unseen real machine data.
The novelty in this thesis lies in the implementation of a condition based monitoring system in the dairy industry, in the decision making technique used to assess the health of the machine, and that in the decision made on the health of the machine occurs after the data has been statistically analysed.
During the initial data collection and algorithm development phase of the project 18 potential major breakdowns were identified and avoided, a saving of £3.6 M to the dairies. During the final stages of the project using the DRSA algorithms within the automatic software, a further 4 faults that could have led to major breakdowns were identified.
Date of Award30 Sept 2016
Original languageEnglish
SupervisorDavid Brown (Supervisor) & Edward Smart (Supervisor)

Cite this