The term augmented intelligence identifies a new management approach that synergistically uses data from the Industrial Internet of Things (IIoT), the new capabilities offered by artificial intelligence and the typical capabilities of human intelligence. In fact, augmented intelligence exploits the strengths of each element mentioned above, namely the ability of the IIoT to acquire countless data collected in real-time, the ability of AI to process them quickly and the human attitude to bring them back into complex situations, creating the so-called “learning feedback”. A paradigmatic example of the advantages that can be found in the application of augmented intelligence tools in the industrial world is that of predictive maintenance. In fact, maintenance is one of the fundamental processes for manufacturing companies, as it affects the quality of production, the duration of machinery over time and the safety of operators. It is a complex process, which is based on the knowledge of the systems and the experience of the staff. It is traditionally practiced “by failure” or “preventively”. To intervene “in case of failure” means to intervene “in emergency”, with loss of production time, possible irreversible damage to the machinery, to the processed products and potential safety risks for operators. For the most important and critical machinery, therefore, it is preferred to operate in a “preventive” manner, basing the action on failure statistics of the machinery components; this method takes a long time and it is difficult to develop as it requires an historical data and extensive experience that is not always available.
With the advent of Industry 4.0, the amount of data collected by machinery began to increase significantly; however, these are still not very usable for maintenance, since they are often designed for the management of production only, they appear as “scattered” in different data lakes and have a reduced history. Holonix has experimented with overcoming these problems on several real cases in the last year and a half, working within the European project Z-BRE4K (www.z-bre4k.eu – G.A. n. 768869).
In the partner manufacturing companies, the PLCs of the machines have been enabled in order to take advantage of the sensors already present inside them to acquire data that are then sent to the cloudi-Live Machines, throughan IoT gateway. The data is then stored, processed in real-time for a pre-analysis and made available to both operators and algorithms. The initial difficulty related to data scarcity was easily overcome by using different AI, performed recursively, so as to generate a first self-training of the system and make it quickly usable for users. Once trained, in fact, the AI processes the data in real-time in order to interpret the weak signals coming from multiple sensors, thus generating prognostic and diagnostic alarms of the machine, alerting of a possible incipient failure, providing an estimate remaining time before failure and suggesting the component to replace. The competence of the maintenance technicians is therefore exploited to evaluate the machine and identify the actual or incipient failure: information then used for the continuous training of the system. The approach, successfully tested within the Z-BRE4K project, has been patented and brings numerous advantages: learning is rapid and requires little time to start providing the first useful results, the ROI is therefore increased and the whole production system more profitable. All this allows you to implement predictive maintenance processes quickly and leveraging the unique knowledge and skills of its experts.
Read the article published in the online Data Manager magazine: click here