Maintenance 4.0: thinking through your project for an optimised ROI5 May 2022
Increased reliability, reduced maintenance costs, optimised planning… As we know, the benefits of predictive maintenance are numerous. But where to start? What are the prerequisites to start your project? Which machines should be targeted first and which mistakes should be avoided?
DiagRAMS gives you concrete levers for a successful predictive maintenance strategy.
Predictive maintenance: what objectives?
Above all, one of the keys to a successful predictive maintenance project is to clearly identify the objective. Is the approach technology-centric or is it based on an explicit business need?
If the project aims to explore the performance of a technology without a clearly identified goal, thinking that the more data there is, the more likely the project is to succeed, then the project is based on the belief that big data is “magic”, which is not the case!
On the contrary, the project set up must solve a specific business problem. The objectives can be multiple: to detect and anticipate targeted failure modes, to diagnose a problem, to identify contexts that generate dysfunctions, to anticipate the crossing of a threshold, etc. It is important to define precisely the subject and the benefits that are expected in the short, medium and long term.
Starting your project: which equipment to target first?
The latest machines, well equipped with sensors and connected? Yes, but…
Should you start your project on a machine that easily transmits information? Yes, this seems easier, but beware of recent equipment that has very few malfunctions and on which no quick return on investment will be possible.
The key is to focus on machines that frequently experience similar malfunctions or require frequent preventive maintenance and that can feed back the data.
Identify critical assets
To ensure a good return on investment, it is important to focus on machines with critical malfunctions or failures, i.e. machines classified as “vital” or “important” under the VIS approach.
Consider the following cases:
- Bottleneck machines on a production line that present a risk of major breakdown with spare parts that are difficult to source;
- Machines with frequent breakdowns/malfunctions whose absence has a strong impact on production volume;
- Machines generating large and costly rejects in case of failure/malfunction.
Depending on the criticality of the equipment, the objectives may vary. In this case, the aim will be to detect operating anomalies, monitor drifts and predict the crossing of safety thresholds.
In other cases considered less critical, the objectives will be different with the analysis of the contexts favouring the good functioning and the dysfunctions in particular for:
- Machines with long set-up/start-up times and scrap creation (e.g. plastic injection machines);
- Machines whose absence has an impact on the volume of production and which have efficiency problems with frequent brief stoppages linked to the context of machine use and requiring operator intervention.
What is the scope of the project?
Starting a predictive maintenance project on a limited scope of machines allows to focus on a precise objective, to involve the teams within a defined framework and to demonstrate the value of the tested solution in the field.
However, it should be borne in mind that the long-term objective is to be able to deploy the solution on a larger scale. It is therefore necessary to assess how to define a scope that is neither too large nor too small to demonstrate both performance and the ability to scale.
Building your dataset: which ones to choose?
It is not the quantity but the quality of the data that matters. To be successful in a predictive maintenance project, it is necessary to ensure that the data contains the right information for the right purpose.
And the frequency of the data will very often condition the wealth of information. It is indeed essential that the step between the values returned is adapted to capture the physical phenomenon. And of course this step will vary according to the machine, its possible cycle, but also according to the physical measurement (temperature, vibration, etc.).
To do this, it is necessary to favour tags rich in information and to select them according to the physical measurement: average over the last second, Root Mean Square (RMS) over the last second, vibratory descriptors (vibratory level per frequency band, or even vibratory descriptor specific to a malfunction). They thus make it possible to reduce the amount of data to be returned while preserving a wealth of information.
Having a good data collection strategy therefore means taking into account not only how a machine works but also the specifics of the physical measurements that must be used to monitor the machine.
Keys to success: what prerequisites?
In order to carry out the project successfully, technical, HR and organisational requirements must be taken into account.
Access to machine data:
Predictive maintenance means monitoring equipment through data analysis. Firstly, it is necessary to ensure that data is available or that the equipment can be connected so that all relevant data can be collected.
Consideration of the context in which the equipment is used:
Predictive maintenance takes into account the context and use of the equipment, unlike preventive maintenance, which only focuses on the equipment itself. It is therefore important to be able to access data from supervision systems (MES, SCADA, etc.).
What’s in it for me? To take into account the variety of contexts in which equipment is used over time (variety of OF, raw materials, cycles, etc.).
A predictive maintenance project involves various teams in an industrial company: management, operations, maintenance, IT department, etc., with issues related to security, data access, solution ergonomics, ROI, etc.
Maintenance & Methods Department:
The maintenance and methods teams have a key role to play in discussing with your predictive maintenance partner and identifying the targeted problems.
Which costs should be given priority (availability, repair costs, scrap, etc.)? Are malfunctions or breakdowns frequent or rare but very costly? Are the machines already monitored? If so, which functional groups are monitored? Which sensors can potentially inform about malfunctions? These are questions that can be answered by the equipment experts involved in the project.
No predictive maintenance without data. But you still need to have access to it! In this sense, it is important to coordinate with the IT department to be able to ensure the exchange of data in compliance with good IT security practices in accordance with the Information System Security Policy.
Another essential point for a successful predictive maintenance strategy is the establishment of a culture of record keeping and reporting.
Machine learning technologies continue to learn over time, so user feedback is essential to improve detection, diagnosis and prediction performance.
Questions about our technology? A predictive maintenance project?
Do not hesitate to contact our team of data science and industry