When we talk about I4.0 we often discuss about data collection, communication infrastructures and security. In this article we will talk about methodologies of management and processing of the collected data. An essential difference, coming from the introduction of the I4.0 technologies, is related to the amount of information collected that allows the use of Artificial Intelligence methodologies and predictive statistical models allowing to carry out complex multi-variable analyzes, not possible with traditional methods.
by Gruppo Meccatronica di ANIE Automazione
The advent of Industry 4.0, together with the support infrastructures has made available a large amount of data, allowing to introduce “Big Data” analysis technologies in an industrial environment. With the term Industrial Analytics we mean, therefore, the use of predictive methodologies and Artificial Intelligence applied to the data collected interacting with sensors at the field level, and with Information Technology and Cloud systems at the management level.
A correct definition of “Machine Learning”
Here is a definition of Machine Learning, in order to clarify its functionality and potential: “it is an Artificial Intelligence methodology that allows computers learning from data, without being explicitly programmed for a specific task”. Everything is done starting from a series of data whose output is known, and with which you calculate the parameters of the used mathematical models. This allows information to be analyzed while the machine is running and to generate supports with high added value for users.
Two main groups of typical applications
Typical applications can be divided into two main groups: prediction and classification of events. In the first group fall the forecast of events and the product quality prediction, in the second the detection and classification of anomalies. Let’s analyze each of the four categories.
1. Anomalies detection
Conditions that deviate from reference behavior are identified, thus they are recorded during “normal” operation. Generally, in industrial applications, the anomaly is an indication of a problem in progress.
This technique allows the detection of critical states that with other ways would not be found, the simplification of complex problems, consequently allows to react in a time such that the problem does not propagate or does not become a “catastrophic” event.
2. Anomalies classification
The found anomalous situations are assigned to typologies of errors defined in the data series and therefore already cataloged. The model can be updated with new sets of information, so its performance can improve over time. This technique facilitates the maintenance work in the plant, as there are indications about the error, above all the reduction in machine downtime when not so frequent or never encountered errors are generated.
3. Event prediction
This technique predicts the remaining time before the occurrence of an event. Typically the residual life of a component or the generation of an error. This facilitates the plant maintenance work, improving the planning of resources, spare parts and operators.
It is important to point out that predictive maintenance analyzes data related to the machine on which it is applied considering the real plant use conditions this allows to prevent unexpected breakages and to use the object of the analysis throughout its life. With preventive maintenance nominal conditions are used, and this is not possible.
4. Product quality prediction
This technique predicts the quality of the product. A comparison is made between a set of records, which define the reference quality, and the data of the work in progress. By detecting the differences, the quality of the final product is predicted and alarms or warnings can be generated.
It also allows the production quality to be monitored continuously, using a large number of variables to reduce waste, given the timeliness in identifying deviation trends with respect to the reference.
Economic and technical benefits of the technology
In order to evaluate a technology, both the economic and technical advantages of all the parties involved must be considered. In our case, End Users and OEMs.
The quantification of machine downtime costs in a production line must include live and induced costs, first of all the decrease in line productivity and the cost of idle operators. This increases costs even by several orders of magnitude compared to the cost of simple replacement. Preventing the occurrence of events presents a considerable economic advantage for the final user of the machines in terms of programming maintenance and production capacity, allowing them to concentrate on their primary objective: production. For OEMs there is an improvement in machine performance and the possibility of offering additional services to customers for whom cost/benefit ratios are positive.