Growing numbers of companies are using the potential of IoT and Big Data to anticipate wear and tear and mechanical malfunction of their equipment. This predictive maintenance process is becoming increasingly accurate thanks to machine learning capabilities.
Whether for controlling machines remotely, monitoring their operation or simulating production processes, many manufacturers are now turning to the Internet of Things (IoT). Already an indispensable technology in Industry 4.0, IoT enables communication with a great variety of objects, from fork-lift trucks to chemical sensors.
How does it work?
The installation of connected sensors in an analytical program enables the constant monitoring of a component, a piece of machinery or a system. How? In various ways, including measuring temperature through infrared images, airflow pressure, or vibration frequency.
The data collected by these sensors is analysed to define a machine’s standard operation. An anomaly is detected by comparison with the benchmark operation. If an anomaly is detected, the maintenance agents are alerted and as such can possibly intervene before the machine breaks down.
Better prediction for greater savings
To boost their ability to anticipate wear and tear, more and more companies are now using the combined potential of IoT and Big Data. For companies – because equipment outages are expensive – the key points are real-time monitoring of machine performance and significant savings.
A survey conducted in three European countries and the United States in 2017 showed that 70 percent of 450 IT decision-makers and on-site service managers do not know exactly when their equipment needs to be maintained or upgraded. Forty six percent of their machines’ unplanned outages are due to component failures. And what are the consequences? Unexpected shutdowns, lasting 4 hours and costing 2 million dollars on average, had a significant impact on production, IT and customer services.
This affects every sector!
Predictive maintenance, the spearhead of the “connected plant” for detecting potential equipment malfunctions on assembly lines, is now being used in all spheres of activity. For example, elevator manufacturer Kone has set up a partnership with IBM to fit its elevators with sensors to detect and anticipate malfunctions. Operating data, stored on IBM cloud servers, are processed by the cognitive informatics of an artificial intelligence program called Watson.
Meanwhile, aeronautical subcontractor Figeac Aero has been deploying predictive maintenance for the last five years on a specific area of data received from its manufacturing equipment: tool vibrations, geometric defects, and clamping strength. This has resulted in the prevention of 40 percent of malfunctions.
Increasingly accurate predictions
Over time, algorithms can be used to create malfunction flowcharts based on fault logs. These models help recognise and then predict potential future malfunctions. Machine learning technologies will progressively enrich these models to enhance reliability and detect all types of faults earlier and earlier.
Dominique Le Beuz, Head of Mobile Network Operations at Orange France, said: “We install sensors at the interfaces of the items of equipment, which form the mobile network to capture and analyse traffic data. We can thereby monitor service quality, as real-time alerts are triggered when network behaviour changes. In particular, this provides assurance that new equipment or new functionalities have been correctly integrated. The sensors provide good visibility over both the network’s service quality and customer experience.”
In addition to service quality, the sensors have other uses. For example, the data they report can feed into third-party applications and offer our customers high-added-value services, for example the “Welcome” text messages when a subscriber arrives in another country and location-based services such as Flux Vision.
Aurélie Piètre-Cambacédès, responsible for the Operational Skill Centre and Network Operations, Orange France, said: “Sensors enable network malfunctions to be anticipated, they check whether the traffic is being handled correctly by comparison with a benchmark behaviour and they can detect the slightest anomaly. Equipment operators can then investigate any issues before the customer is affected. When a customer has a problem, analysing how the call was handled across the network can sometimes help us anticipate and prevent much bigger, wider problems.”
What’s more, this ability to anticipate has a clear financial impact on the allocation of technical resources and on crisis management costs if a problem arises – all representing costs for our corporate clients. With the advent of automatisation in network operation and the installation of application programming interfaces (API) on sensor systems, the operational gains could potentially be even more significant for Orange than for our customers.
These sensors being deployed on the France mobile network, Orange will soon be able to correlate data transversally and end-to-end. This is becoming indispensable to optimise analysis of how we handle very complex incidents because the network densification sometimes makes it difficult to pinpoint the incident. All our subsidiaries with a mobile network can benefit from this system of sensors, and we also provide support for its installation and operation.
