Prescriptive Maintenance (RxM)

What is prescriptive maintenance (RxM)?

The easiest way to explain prescriptive maintenance (often abbreviated as RxM) is to correlate it with predictive maintenance, as it is a natural extension of it.

Predictive maintenance is a term you should already be somewhat familiar with. It represents the usage of sensors, IIoT, and big data to generate data-intensive, condition-based predictions of asset health and performance. Advanced analytics are used in near real-time to identify reliability issues that could impact operating efficiency, alerting the maintenance team to schedule preventive/corrective actions to maintain/restore optimal operation.

Basically, we have algorithms that assess continuous data streams and compare them against equipment models formed from historical performance information. The software seeks recognizable patterns, and if discrepancies from established norms are identified, the emerging issues are analyzed. 

This is enough to predict equipment failures. But prescriptive maintenance goes a step further. 

It uses custom-built prescriptive algorithms to process the potential outcome of multiple scenarios. The result? A list of possible solutions, ranked based on predetermined factors you defined while building the algorithm.

If you let it, prescriptive analytics can autonomously apply the most efficient solution to alleviate or mitigate the identified problem. These solutions may include everything from automatically modulating the operation of a specific piece of equipment to alerting maintenance technicians and operating personnel about the specific actions they need to take to maintain optimal equipment efficiency. 

Dan Miklovic from LNS research explained it well in his post: 

Let’s say a piece of equipment is showing increasing bearing temperature. Predictive analytics looks at the temperature profile and tells you it is likely to fail in X amount of time. On the other hand, prescriptive analytics tells you that if you slow the equipment down by Y%, the time to failure can be doubled, putting you within the already scheduled maintenance window and revealing whether you can still meet planned production requirements.

Already in use in the aerospace, oil, chemical, and mining industries, prescriptive technology promises a competitive advantage to manufacturers in any asset-intensive industry where equipment uptime and efficiency is paramount.

Table of Contents

Pros and cons of prescriptive maintenance
Real-world uses of prescriptive maintenance
Steps to implement prescriptive maintenance
Not feeling ready for prescriptive maintenance?

Pros and cons of prescriptive maintenance

There isn’t a single maintenance strategy that is a perfect solution for every asset. While it offers amazing benefits, prescriptive maintenance comes with a distinct set of challenges.

Benefits of using prescriptive maintenance:

• Provides multiple scenario modeling and uses machine learning and artificial intelligence to prescribe and automate a broad range of options and their respective outcomes.
• Assists businesses in understanding the economics of specific actions proposed.
• Optimizes maintenance operations, minimizes downtime (unplanned downtime as well), and increases operating efficiency.
• Creates a digital model of equipment, allowing simulations of future additions or modifications before incurring a capital expense.
• Can be used to solve difficult equipment issues such as multiple or cascading failure modes or almost imperceptible equipment deterioration over time.

Drawbacks of implementing prescriptive maintenance:

• The costs of retrofitting the equipment with sensors and implementing required artificial intelligence (AI) and machine learning (ML) capabilities.
• The system is not immediately effective, with equipment failures possibly required to teach the machine learning platform, which can put additional strain on asset management.
• Regulation will take some time to catch up to the technology, possibly requiring mandated replacements despite prescriptive system solutions suggesting otherwise.
• As an emerging technology, prescriptive maintenance will need to mature to generate confidence in its unsupervised management of safety-critical systems.
• Workplace culture can delay or disrupt implementation, requiring considerable attention to early communication, consultation, and education.

Long story short, prescriptive maintenance is very powerful, but also very expensive. It is currently viable only for the most critical pieces of equipment that have huge repair or replacement costs.

Real-world uses of prescriptive maintenance

Far from being theoretical concepts, prescriptive analytics and prescriptive maintenance are in use now, worldwide, by companies seeking greater operating efficiency. Here are some examples of different industry uses and realized benefits.

RxM example #1: Airplane maintenance

Aerospace has quickly realized the potential savings from prescriptive maintenance, with Boeing offering data analytics services to clients. Their Boeing Analytx team has used data analytics to diagnose early failure modes in the Boeing 787 hydraulic spoiler power control unit (PCU). 

The PCU has two internal motor windings, whose dual failure signals an alert on the flight deck and grounds the aircraft until spares are sourced and unit replacement occurs. While an uncommon event, it is extremely disruptive to airline operations when it does happen.

RxM example #2: Elevator maintenance

The German multinational conglomerate ThyssenKrupp utilizes prescriptive analytics in its elevator servicing operation. 

Machine learning uses the data from its elevator network to identify tasks critical to the safe and reliable operation of the elevator. It can now predict five days in advance when an elevator will shut down due to door problems, automatically scheduling a technician and providing the four most likely actions to alleviate the pending issue.

Such accuracy prevents outages with passengers onboard and allows technicians to fix the problem with 90% accuracy.

RxM example #3: Pump maintenance and repair

A multinational mining company uses autonomous analytic systems on circuits critical to their metal-refining process and receives specific pump repair instructions 40 days prior to the actual pump failure.

The Boeing team uploaded their global fleet knowledge and used data scientists and design engineers to create a predictive algorithm to give advance notice of failure. 

It was only a short step from there to the system taking autonomous action within user-defined constraints. Prescriptive detailed work instructions are sent to the destination outlining the access, spares and technical skills required to affect component replacement during refuel and turnaround, minimizing downtime and maintaining strict schedules.

Steps to implement prescriptive maintenance

The path to an operating prescriptive maintenance system is similar to establishing a predictive maintenance program, just with some extra steps.

Here’s a basic outline of the implementation process so you understand the scope and amount of work that needs to be done.

Step 1: Plan out the implementation

The planning phase requires you to understand and communicate your expected outcomes. Given the costs and complexity of implementation, you should target mission-critical business assets with high capital or repair costs. 

When first implementing such a system, it is wise to start using one piece of equipment to prove system operation before rolling it out to the rest of the plant. In other words, it’s recommended to start with a pilot project.

Step 2: Collect historical data

Developing an accurate prescriptive algorithm depends on the amount of relevant data available. Historical data for the equipment may come from hard-copy maintenance records or digital databases from your computerized maintenance management system (CMMS). 

This asset performance data, coupled with technician and engineering insight, will form the genesis of the failure mode analysis, as well as the model upon which the algorithms will carry out its future analysis.

Step 3: Perform Failure mode effects and criticality analysis (FMECA)

When building a digital model of equipment, you must:

• Identify how equipment has failed in the past
• The effects of those failures
• The level of criticality and risk involved in those breakdowns 

This study will inform the autonomous software system’s predictive and prescriptive aspects. Here is an article that provides a detailed discussion on how to carry out an FMECA study.

Many maintenance teams will complement their FMECA analysis with an RCM analysis. When building predictive or prescriptive algorithms, the more data sources you can pull from, the more accurate and useful the algorithms will be.

Step 4: Establish data capture systems

The heart of prescriptive maintenance is the real-time data feed from the equipment using sensors and software. Either hard-wired or via wifi, equipment condition signals pass through a network where sophisticated monitoring software captures, aggregates, and analyzes the data. 

With the rapid technological advances introduced by the industrial internet of things (IIOT), modern condition-monitoring sensors are digitized, miniaturized, built for operation in harsh environments, and have low power draws even while communicating over secure wireless systems.

Such systems must be robust and have sufficient bandwidth to cope with the large volume of data processed. Sensor selection and data monitoring points will be informed by the outcome of the FMECA process carried out in the previous step.

Step 5: Build your predictive maintenance algorithm

At this point, you piece together all the components to form an equation that can – based on its data analysis – predict each of the failure modes identified earlier. 

The initial input for the algorithm will usually include:

• Original equipment manufacturer (OEM) information and recommendations
• In-service history pulled from hard copies or digital equipment maintenance logs
• Engineering, design, and maintenance technician input
• Data from any kind of non-destructive testing you might be running regularly (in some cases, the available data from destructive testing could also be helpful)
• Machine monitoring data captured from real-time sensors

A data scientist will create the initial models before integrating machine learning and artificial intelligence platforms to begin the process of learning and refining the model.

Step 6: Incorporate prescriptive actions 

Responses to predicted failures will vary depending on the specific issue’s immediacy, mode, and criticality. They may also be multi-faceted. The prescribed actions might encompass automatically altering an operational state or restricting speeds, pressures, or temperatures to reach planned maintenance windows. 

At the same time, the system may alert maintenance technicians to plan for a component change, highlighting the procedures to follow and the time required to effect the change. The system can prompt the store’s department to prepare the appropriate spares, and resource scheduling can be alerted about the need for specific skills rostering at the next maintenance shut.

These initial actions must be programmed into the algorithm based on input and guidelines from the various departments with knowledge of the correct processes.

Step 7: Roll-out the system and let it learn

Roll out the system on the chosen equipment. The initial algorithms will not be perfect. You will have to allow the machine learning and AI platforms to learn and improve. The analytics platform will monitor equipment operation and analyze how the prescribed operational changes or maintenance tasks affected equipment performance, its operating condition and planned production. 

The base model will need to be refined as more data starts coming in. It might take a while until you are confident enough in the algorithm to let it make important decisions without significant human oversight.

Step 8: Work on culture and change management

Prescriptive maintenance is a great tool for optimizing your maintenance resources and creating an efficient workplace. However, you will fail to achieve complete success if all stakeholders (operators, maintainers, and supporting functions) are not onboard with using the technology. 

Not only must internal processes and procedures change to support the new way of working, so too must the actions of the individuals involved. 

A comprehensive change management process is required to ensure the initiative is accepted throughout the organization. Appoint project sponsors to support frequent and comprehensive communication. Amend maintenance practices, roles and responsibilities to assist employees in effectively acquitting their roles. 

Remember: If the new system makes people’s jobs less fulfilling or more difficult,  people will not want to use it – and your expensive initiative is likely to fail.