How to implement a Predictive Maintenance strategy
by Frederick Mannings, Data Engineering Lead
How to Implement a Predictive Maintenance Strategy
After reading all the above benefits about predictive maintenance, you may be asking yourself how to implement it in your company. There are many ways for companies to realise value from PdM. But what we have done is outlined a broad strategy of implementation that is applicable to many case studies. Here is the general guideline to implementation.
Understanding Business Objectives
On what system(s) do you want to implement predictive maintenance? Is it an entire factory or just a machine, or maybe even a subsystem of a machine? What is the expected benefit? What KPIs are you going to monitor? What is your budget? What is your timeline?
These are all important questions to consider when setting out on building a PdM strategy. Having clear visibility on what the value proposition means to your business is critical in approaching a predictive maintenance solution from the right direction.
Engineering Requirements
Analyse the existing system interface, Define new requirements and bring it down to monetary savings. Define what cost savings are feasible, and necessary to help steer your predictive maintenance capability. In doing so, you will confirm/discover your current monetary expenditure of service and maintenance.
Validation and Verification
It's essential to know how you are going to confirm that your requirements are what you need and that your solution meets them. Mapping out a VnV procedure is critical to achieving this. For each requirement, confirm that these are going to meet your business need, and determine a strategy to assure that your solution satisfies the requirement.
Solution Design
Define what data is going to be collected, what you are going to do with it. Sometimes, the infrastructure to access all the data you need does not exist in the first place.
Data Collection
Up until now you have defined the objectives and requirements of your predictive maintenance strategy. In doing so you have also specified the targeted fleet and the associate required data. Aggregating system data is the next critical step in the process.
Creating Conditional Indicators
Interpret the raw data to create conditional indicators. The conditional indicators should correlate with system performance. Condition indicators may be an aggregation at time intervals; e.g. cycle, day, unit of work.
Building RUL Models
Build Remaining Useful Life models from the conditional indicators. If an amount of historical data has been collected, a similarity model can be constructed. Otherwise, generate a lifetime model utilising domain knowledge and/or first principles.
Verification and Uncertainty Quantification
Given your remaining useful life model, apply the verification methods you have previously defined. Additionally, it's at this stage that you would quantify any uncertainties in your model.
Deployment and Monitoring
It's now time to deploy your models into production. You need the infrastructure to determine the condition indicators in near real time if not real time, along with the propagation of your remaining useful life model.
The expertise necessary to execute the above steps spans many domains of engineering, project management and data science. Prior to building a PdM strategy, it is worthwhile considering the resources available to you.
When it comes to maintenance, most companies are still stuck at using old maintenance strategies such as reactive or preventative maintenance. They do not always realise it, but they are missing out on a huge opportunity to reduce costs and downtime while maximising customer satisfaction, all of this thanks to predictive maintenance.