Artificial intelligence for maintenance

12 April 2022

Artificial Intelligence (AI) techniques can be used to provide solutions for monitoring the operating status of industrial equipment. These solutions can both detect phenomena that have already occurred and identify others that are new. Interview with Jean-François Bouin, president and co-founder of DiagRAMS Technologies, which offers a predictive maintenance software solution for the early detection of anomalies.

 

Why is artificial intelligence suitable for predictive maintenance?

JEAN-FRANÇOIS BOUIN:

Industry is a complex environment with heterogeneous machinery, high production variability and if some breakdowns are recurrent and known, many are new!

It is therefore necessary to be able to deploy artificial intelligence (AI) solutions capable of identifying new abnormal behaviours and diagnosing new types of malfunctions. This is precisely what we offer at DiagRAMS. Our predictive maintenance solution allows early detection of anomalies with an unsupervised approach, without seeking to recognise abnormal operation but seeking to detect that the operation is simply unusual.

The aim is to monitor the health of the equipment, to detect early warning signs of all types of failures and to quickly inform the maintenance teams of a possible problem so that they can intervene at the right time.

In this context, AI is a formidable ally for operational staff, who thus spend less time solving urgent problems and regain serenity and time to work on higher value-added tasks such as making machines reliable.

 

How do AI techniques differ from conventional computer processing?

JEAN-FRANÇOIS BOUIN:

Traditional approaches aim to detect events based on phenomena and rules known by experts in the field, such as alerting when a threshold is exceeded on a simple sensor.

These approaches are limited because they are fixed in time and can be time consuming because they require the translation of expert knowledge into specific rules. However, this is not sufficient when one wants to follow the drifts of functioning over time, understand the evolutions or discover determining elements on a wider and more varied spectrum of parameters that may be completely new.

This is precisely what AI techniques offer, as they will seek to draw on a history of data to learn and identify what seems normal or abnormal in the operation of a machine or process.

Furthermore, AI techniques have learning capabilities that improve their performance over time. They are not based on fixed rules but on models that will evolve over time.

 

What are the advantages of AI for maintenance?

JEAN-FRANÇOIS BOUIN:

AI techniques can be used to monitor industrial equipment in real time, detecting phenomena and behaviours that have not occurred before and identifying some that are new. This is because their learning capabilities can distinguish between correct, incorrect or simply unusual operation of equipment.

Thanks to the analysis of a variety of parameters, they can detect certain weak signals announcing breakdowns and predict how things may develop. This makes it possible to plan maintenance interventions, mobilise technical staff who will intervene at the right time and anticipate the purchase of any spare parts.

To be effective, AI techniques must respond to the specificities and challenges of the industrial context: few available failure histories, variety of equipment use contexts, multi-source, multi-format data, etc. This is why DiagRAMS AI is the result of several years of R&D in the Inria research centre in order to meet these challenges and offer a turnkey solution for maintenance teams who need to optimise maintenance interventions, anticipate the purchase of possible spare parts and reduce emergency interventions.

AI also helps to avoid wasting raw materials and energy in order to reduce the environmental impact of production tools.

 

How can equipment be monitored with AI-based tools?

JEAN-FRANÇOIS BOUIN:

The machine must be connected in order to provide information on its operating status and the values of key parameters (flow, temperature, pressure, vibration, current, etc.). It is also necessary to be able to transmit information on events occurring during its operation (maintenance interventions).

On relatively new machines, it is usually possible to make do with the available data. In some cases, additional sensors may be required. In any case, it is important to consider what data is relevant to the machine before deciding to install additional sensors, and it is best to study the nature of the data carefully so that only data of real value is taken into account.

When several industrial machines and equipment are interconnected on a production line, it makes sense to monitor all components, up to and including the conveyor systems. However, when starting a project, the first focus is usually on the so-called critical machines, where a malfunction leads to a lot of rejects and financial losses. It is on this type of machine that the hope of gain is greatest when implementing an AI-based monitoring system.

DiagRAMS software

 

What are the obstacles to the implementation of AI-based maintenance applications?

JEAN-FRANÇOIS BOUIN:

Estimating the financial losses due to the stoppage or malfunction of a machine is based on many parameters: increase in scrap, price of raw material, cost of spare parts, delay in delivery, deterioration in quality, lengthening of maintenance intervention times, etc. Manufacturers need to identify the expected ROI on their equipment.

This is why we work with our clients to identify the limited perimeter on which there is a clearly identified hope of gain. This allows us to gauge the effectiveness of the solution and its impact on the organisation of the company implementing it.

Once the whole process and considerations are well understood, the scope of monitoring is extended to other equipment. It is not necessary to start with those that are easiest to connect but with those that offer a clearly identified hope of gain.

For us, an AI-based predictive maintenance monitoring system must be a decision support tool. The prediction process must be clearly explained to the person who will be making the decisions.

AI techniques must provide qualified information to complement traditional operational data. If they only deliver information that all the technicians in a factory already know, they are of no interest.

 

How to implement an AI-based application for maintenance?

JEAN-FRANÇOIS BOUIN:

We start by discussing with the customer the scope of the machines to be monitored and the type of malfunction they wish to identify. We ensure that the machines have the appropriate sensors and that they can provide the data required for monitoring. The information to be collected and its frequency of acquisition must be determined with the help of the customer’s experts who have a good knowledge of their industrial processes. Industrial communication gateways are then set up on the machines to select, extract, format and send the data to our analysis platform accessible in Saas mode.

We configure our software to meet the process and data specifics and application requirements. We select the most appropriate AI algorithm and data analysis method for the situation and set up the automated monitoring process.

 

What are the priorities in their deployment?

JEAN-FRANÇOIS BOUIN:

Maintenance technicians who know their machines and operating processes inside out should of course be involved in the preparation, development and deployment of AI-based predictive maintenance applications. They need to be made aware that such a solution will save them time. The number of their interventions for emergency repairs will decrease. This will allow them to focus more on making the machines more reliable.

But the success of such a project requires the cooperation and involvement of the different structures of the company, from the management to the IT systems managers, starting of course with the maintenance department.

 

First published on: J’automatise by Youssef Belgnaoui

Switch to
predictive maintenance !