An Intelligent Approach to Optimal Predictive Maintenance Strategy … · 2013. 6. 21. · maintenance, i.e. - determination of a machine condition while in operation. • In the

1http://www.iict.bas.bg/acomin

6/21/2013

INSTITUTE OF INFORMATION AND COMMUNICATION TECHNOLOGIESBULGARIAN ACADEMY OF SCIENCE

I. Mustakerov, D. Borissova

An Intelligent Approach to Optimal Predictive Maintenance Strategy Defining

AComIn: Advanced Computing for Innovation

IEEE – International Symposium on INnovations in Intelligent SysTems and Applications INISTA'2013, June 19-21

6/21/20132

http://www.iict.bas.bg

Contents

• INTRODUCTION

• PROBLEM OF OPTIMUM MAINTENANCE STRATEGY

• GENERALIZED METHODOLOGY FOR PREDICTIVE MAINTENANCE

• MATHEMATICAL MODEL FORMULATION

– Optimum Maintenance Strategy Defining

– Decision Making Algorithm

• ILLUSTRATIVE NUMERICAL EXAMPLE

• CONCLUSION


IEEE – International Symposium on INnovations in Intelligent SysTems and Applications INISTA'2013

6/21/20133


Introduction

• Technological development resulted in increased complexity in bothindustrial machinery and production systems. The modern industry isconstantly demanding for work at high reliability, low environmentalrisks, and human safety while operating their processes at maximumyield. Therefore, prevention of failures and early detection ofincipient machine and systems problems increase the usefuloperating life of plant machinery.

• Fault detection and diagnosis in the early stages of damage isnecessary to prevent malfunctioning and failure during operation.This will reflect in substantial benefits achieved through the use ofoptimization techniques in plant operations by improving theresource utilization at different levels of decision-making process.



6/21/20134


Introduction

• The need for effective maintenance planning is constantly increasing. Itis recognized that maintenance planning is complex problem involvingnot only condition monitoring but also expert knowledge. One majorimprovement in maintenance technology is using of the predictivemaintenance, i.e. - determination of a machine condition while inoperation.

• In the paper, an intelligent approach to optimal predictivemaintenance strategy defining is proposed. It is based on methodologyfor predictive maintenance that increases reliability by determining theoptimal maintenance strategy. For the goal, an algorithm based oncost-benefit analysis and optimization tasks solution is developed.



6/21/20135


Introduction

• The premise of condition based maintenance is that regularmonitoring of the actual mechanical condition of equipment andoperating efficiency of process systems will ensure the maximuminterval between repairs; minimize the number and cost ofunscheduled outages created by machine failures and improve theoverall availability of operating plants.

• Condition based maintenance techniques rely on assessment of thesystem’s condition, based on data collected from the system bycontinuous monitoring. The goal is to determine the requiredmaintenance plan prior to any failure.

• The maintenance strategies aim to minimize the costs byimprovement of the operational safety and reduce the severity andnumber of in-service system failures.



6/21/20136


Problem of Optimum Maintenance Strategy

• To provide optimal decision making on when to take equipmentdown for maintenance should be integrated with overallmanufacturing operations. This is why additional information aboutsensing and data acquisition is needed. This ensures accurate data foravailable signals from the equipment to be collected. These signalsare captured using specific sensors.

• The condition monitoring and establishing of health assessment is abaseline for equipment performance as critical step in enablingpredictive maintenance solutions.

• Standard condition monitoring techniques like statistical processcontrol or advanced process control could be employed primarily inprocess monitoring.



6/21/20137



• A maintenance program needs to define different maintenancealternatives for different equipments. Therefore, the maintenanceproblem could be recognized as a combinatorial optimizationproblem. When considering a decision making problem some startingprerequisite should be noted that in all cases there exists a decision-maker (DM) and the optimal solution implies the existence of afunction, which should be optimized.

• Let x be a vector containing p decision variables. Mathematically, an optimization problem can be expressed as:

Minimize/maximize fi(x) for i = (1, 2, …, n) (1)

s.t. gj(x) ≤ 0 for j = (1, 2, …, J) (2)

where x = {x1, x2, …, xp} xi is the i-th decision variable (i-th alternative).



6/21/20138


Predictive Maintenance Methodology

The proposed predictivemaintenance methodology inFig. 1 includes:

1) collecting information aboutthe system condition,

2) defining the system state andgoals and criteria,

3) defining alternatives andformulating of optimizationtask for a decision making,

4) task solution and defining ofthe alternative/s,

5) acceptance of the definedalternative/s

6) interactive influence of theDM during various stages.


Fig. 1


6/21/20139



The algorithm for implemen-tation of stages 6, 7, 8, 9 and10 from the previousdiagram is visualized – Fig. 2.

As a result of optimizationtasks solutions a group orsingle alternative isproposed for evaluationagainst the objectives. Theselected alternative/s shouldbe approved by the DM andif not - optimization taskredefinition are to be done .



Fig. 2

6/21/201310



The cost-benefit estimation can be used to define the most appropriatedecision making strategy. Maximal cost-benefit values define the worstcase scenario i.e. a proper maintenance reaction is required. Two typesof costs-benefit estimations are proposed:

(3)

(4)

where Crepair_component is the cost for component repair, Cnew_component is the costof replacement by new component, α is coefficient indicating how many timesthe component has failed, Premaining is the profit during the componentremaining useful time and β is coefficient that indicates whether thecomponent is repaired (0 < β < 1) or replaced with a new one (β = 1).


remaining

componentrepair

PC

CBE repair

βα _=

remaining

componentnew

PC

CBE new _=


6/21/201311


Decision Making Algorithm

• If an alert alarm from the conditionmonitoring equipment appear, thequestion is what to do? The decisionalgorithm can be realized by two steps.

• On step I the alternative is – to repairor to replace the machine as a whole. Ifthe decision on the first step is toreplace the machine by new one thedecision process ends. If the decisionon step I is to repair machine, next stepII is executed.

• On step II the question – what to dowith each particular machinecomponent – repair or replace? –should be answered.



6/21/201312


Mathematical Model Formulation

• Step I: To repair or to replace the machine as a whole is result of solution ofthe following optimization task:

maximize (5)

s.t. x + y = 1 , x, y ∈{0, 1}• Step II: The repair/replace strategy for each particular component is

defined by a run of the following optimization problem:

maximize (6)

s.t. ∀ i ∈{1,2,…,n}: , xi , yi ∈{0, 1}

is the cost-benefit evaluations for repairing of i-th component,is the cost-benefit evaluations for replacing of i-th component,

x, y are binary integer variables assigned for each alternative


CBEyxCBEn

1i

newi

n

1i

repairi

+∑∑

==

CBEyCBExn

1i

newii

n

1i

repairii

+∑∑

==

∑=

=+n

iii yx

1

1

repairiCBEnewiCBE


6/21/201313


Illustrative Numerical Example

• The proposed approach is illustrated by considering a real data exampleof vibration feeder as object of predictive maintenance. The vibratingfeeder consists of the following components: 1 – vibrating frame,2 – spring, 3 – vibrator, 4 – motor vibrating device and 5 – motor.

• The cost-benefit analysis calculations are useful, because they allowexamining the options and making more informed choices. The vibrationfeeder cost-benefit evaluations of the component repair/replacealternatives corresponding to the estimations (3) and (4) are shown.



6/21/201314



• The optimization problems (5) and (6) are used to formulatecorresponding tasks for step I and step II to define the “best”alternative from a set of available alternatives.

• Step I: The optimal strategy to repair or to replace the machine as awhole takes into account optimization problem (5) involving the cost-benefit evaluations from Table I:

max {x (0.85 + 0.70 + 0.65 + 0.90 + 0.80) + (7)+ y (0.75 + 0.80 + 0.75 + 0.85 + 0.70)}

s.t. x + y = 1, x, y ∈{0, 1}The solution results of optimization task (8) are shown in Table II.



6/21/201315



According solutions results data in Table II the optimal maintenancestrategy is to repair the machine. This means step II should be executed.• Step II: The optimization problem (6) is transformed into following

optimization task:max {(0.85 x1 + 0.75 y1) + (0.70 x2 + 0.80 y2) + (8)

+ (0.65 x3 + 0.75 y3) + (0.90 x4 + 0.85 y4) + (0.80 x5 + 0.70 y5)}s.t. ∀ i ∈{1,2,…,5}: , xi , yi∈{0, 1},

The solution results of optimization task (8) define the optimal strategiesfor each component are shown in Table III.


∑=

=+5

1

1i

ii yx


6/21/201316


CONCLUSION

• The example of vibration feeder is used just for illustration of theproposed approach. In practice, defining of optimal predictivemaintenance decision strategy for other machines consisting of muchmore components will be complex mathematical optimizationproblem. In such cases the proposed intelligent approach to optimalpredictive maintenance strategy defining will be a better alternativeto intuitive decision making.

• The units in cost-benefit analysis are standard money units thus bothcosts and benefits can be compared directly. In some cases, where itis difficult to express the benefits into money cost-effectivenessanalysis could be used as cost-minimization technique. Otherobjective functions could be used in estimation of machinecomponents maintenance strategy.



6/21/201317


CONCLUSION

• The described intelligent approach to optimal predictive maintenancestrategy defining is based on optimal decision making algorithm.

• Different maintenance alternatives – repair or replace, can beevaluated using the concept of costs-benefit analysis by means ofproper optimization tasks formulations.

• Two different kinds of optimization problems are formulated:1) defining of maintenance strategy for machine as a whole;2) defining of maintenance strategy for each particular machinecomponent.

• The formulated optimization tasks are used in predictivemaintenance decision making algorithm included in proposedgeneralized methodology for optimal predictive maintenance.



6/21/201318


Acknowledgment

• The research work reported in the paper is partly supported by the

project AComIn – Advanced Computing for Innovation, grant 316087,

funded by the FP7 Capacity Programme (Research Potential of

Convergence Regions).



An Intelligent Approach to Optimal Predictive Maintenance Strategy … · 2013. 6. 21. · maintenance, i.e. - determination of a machine condition while in operation. • In the

Documents