Research On Methodology Of Intelligent Maintenance Scheduling With Energy Consumption Considered For Parallel-serial Production Systems

Intelligent maintenance scheduling (IMS) methods are developed based onintelligent diagnosis techniques. In IMS,“Condition-Based Maintenance” istaken as the basic strategy, and intelligent methods are utilized to conductdemand-based maintenance on facilities. By conducting IMS, the degradationprocess of facilities can be effectively slowed down, and facilities’ performancecan be kept in high level, thus resulting in high reliability and low energyconsumption, with minimized maintenance cost and production loss caused bymaintenance activities. Therefore, IMS is of great importance for manufacturers.An IMS model for series-parallel production systems is established in thispaper. Based on Weibull distribution, the reliability degradation of facility ismodeled. By improving Malik’s Proportional Age Reduction Model, andintroducing the acceleration factor of reliability degradation, a maintenanceeffect model is established with four types of maintenance actions considered, i.e.minor maintenance, medium maintenance, overhaul and replacement. An“Age-Energy Consumption” model is established with facility’s regular wearcurve and the relation between wear and energy consumption considered. Astructural model for series-parallel systems is established to describe thestructural dependence among facilities. Based on the models above, the system’senergy consumption and reliability are modeled respectively. The maintenancecost model is also presented, which consists of two parts, i.e. direct maintenancecost caused by the utilization of maintenance resources, and the production losscaused by facilities’ downtime. The structural dependence among facilities, thelimited buffer size and maintenance resources is considered as constrains in thecalculation of production loss.Three levels of maintenance scheduling, i.e. decision-making ofmaintenance timing, facility grouping, and the sequencing of maintenance units,are studied in this paper.Multi-Objective Comprehensive Learning Particle Swam Optimization(MOCLPSO) is utilized for decision-making of maintenance timing. Theminimized energy consumption, maintenance cost and maximized reliability ofthe system are set as three goals of the optimization. The procedures ofMOCLPSO are introduced. Key problems in the application of MOCLPSO arediscussed, including the coding of solution, learning process of particles, thesetting of inertia weight and learning factors, and the update of external file. Two comparative methods for multi-objective particles are presented, namely Paretospace comparison and linear weighted comparison.Hierarchical Clustering method is applied in facility grouping. Similaritiesbetween facilities that are related to maintenance, including the similarities inlocation, facility type, type of maintenance action, structural position in thesystem, and maintenance time needed, are considered. The measurements ofthese similarities are also presented. Weighted Average Linkage method isapplied in the clustering process. And the partition of clusters is conducted byreferring to the inconsistency coefficient of among links.A sequencing rule extraction method based on CART for maintenance unitsis proposed. The procedures of CART are introduced. The pattern of sample isdesigned, with some key factors that affect the maintenance costs considered,including facility’s maintenance time, maintenance resources required, andsystem’s productivity reduction caused by the facility’s downtime. In thegeneration of sample sets,20regular sequencing rules are designed and GeneticAlgorithm (GA)–based sequencing method is presented to generate two sets ofgood sequencing solutions respectively. Based on each solution set, one sampleset is generated. The growth and pruning process of decision tree are alsopresented. The rule extraction and the conversion of sequence are designed.A sliding bearing production system is introduced to verify the effectivenessof the methods proposed in this paper. First, the multi-objective decision-makingproblem of maintenance timing is solved by utilizing MOCLPSO based on Paretocomparison and MOCLPSO based on linear weighted comparison respectively.The optimization process is analyzed and the effectiveness and applicationconditions of the two methods are discussed. Second, a solution of the firstproblem is taken as an example to testify the effectiveness of the proposedgrouping method. The application of group maintenance is discussed accordingto the simulation results under different resources constraints. At last, theproposed sequencing rule extraction method is applied. The growth and pruningof CART based on samples generated by regular rules and samples generated byGA are discussed and analyzed. Based on the simulation results of test samplesand the running time of each method, regular rules, GA, rules extracted fromGA-based samples and regular rule-based samples are compared and discussed.