Being able to look into the future – who hasn't wished for that? In business in particular, a lot of money often depends on decisions that must be made on the basis of assumptions. That's why there are several technologies that try to investigate the future based on known data or to make it predictable.

Predictive Analytics – Evaluating and interpolating data

One of these technologies, which is also included in Industry 4.0, is predictive analytics and the predictive maintenance based on it. For example, data from sensors of all important components, bearings and drives of a machine – or ideally an entire park of similar machines – is continuously collected and stored in a so-called data lake. Tools, such as drills or milling cutters, can also be monitored.

Two practical examples of predictive analytics

1. An artificial intelligence (AI) examines this data and information for irregularities and deviations from normal operation. Drills and milling cutters that are becoming blunt, for example, can be detected by increasing cutting forces and thus increasing power consumption of the spindle. 2. Another possibility is to record the vibrations during machining by sensors. If their pattern changes, this may indicate blunt or damaged tool cutting edges. In this way, defects or quality problems can be detected long before a failure or problem occurs. Predictive maintenance makes use of these findings to schedule maintenance in good time before the failure and to carry it out during a break in work, for example. This in turn has a positive effect on maintenance.

What is a data lake?

Data Lakes are data storage solutions that are particularly suitable for large data sets. They can enable completely new real-time analytics. Their highly scalable environment supports extremely large data sets and ingests data and information in their native format from various data sources.

Detecting deviations requires large amounts of data

The deviations that need to be detected for predictive analytics are often only marginal and, moreover, not easy to identify. An AI first learns the "normal state" of the data and then detects deviations from it. It can then also sound the alarm, but it is not clear where the deviation is coming from. Accordingly, false alarms will often occur. In order to be able to identify the cause of a deviation, the AI needs examples, i.e., deviation patterns, for which the reason is known.

How to develop predictive maintenance examples for AI

These can only be generated by continuing to run a machine until, for example, the bearing is defective or the drill is recognizably blunt. Then, based on the progression of the patterns, the AI can also name the cause and assess the importance of the deviation. This predictive maintenance is also important for Industry 4.0, among others. The basic problem is distinguishing between important and unimportant deviation patterns. An example of an unimportant deviation is when temperature fluctuations in a manufacturing hall cause storage temperatures to change with the season. These changes are naturally detected by the AI, but a well-trained AI understands the correlation with the outside temperature and refrains from raising an alarm. If a warehouse temperature differs in its course from this change, the AI sounds an alarm. The example shows: Training an AI to reliably look into the future requires a lot of data, which can be obtained either in long test series or in a large number of machines. A machine manufacturer that collects data from all of its machines that are operated by customers will be able to collect a sufficient amount of data more quickly than a customer that operates only one or a few such machines. Collecting and analyzing machine data therefore brings benefits for all users of such a machine.

Quantity matters: Identifying weak points in the process

Above all, the collection of data across an entire machine series enables the manufacturer to identify weak points in the machine. If the same defect occurs on many machines, this indicates that the component in question is undersized or incorrectly designed. IoT can be used to identify such weak points at an early stage and eliminate them, for example, by replacing them as a gesture of goodwill, even on machines where the defect is still a long way off. Of course, it also makes sense to collect status data within a company. For example, tool service lives can be recorded to the minute and, if a supplier is changed, it is possible to track exactly how the new tools behave in comparison to the tried-and-tested milling cutters or drills.

Legally compliant data collection and analysis

Demanding customers of contract manufacturers in the machining sector require data on machine availability to safeguard their fragile lean production process chains. Here, predictive analytics offers the possibility to automatically extract the relevant data from the accumulated data mountain. On the other hand, predictive maintenance makes it possible to provide delivery guarantees for the future without excessive risk.

Predictive maintenance: intervening before the problem arises

By monitoring the machine parameters, it is possible to detect a coming defect long before it becomes acute. This eliminates the need for preventive maintenance work, such as changing drive belts after a certain number of operating hours. The aim of predictive maintenance is now to identify the need for service early enough to allow time to procure the necessary spare parts, tools and resources and, optimally, to retain enough leeway to schedule maintenance during an off-peak or break period when the machine is not needed. Predictive maintenance enables service models in which the manufacturer monitors the machines and initiates service automatically. This can then be covered by a service flat rate, for example. The user of the machine has the great advantage that he can rely on his machine. It is serviced when there is time and is always in optimum condition.

Service and maintenance: Flexible and targeted instead of rigid

Preventive service, in which parts are replaced at fixed intervals, also means that many components are replaced prematurely. The manufacturer will always measure the service intervals in such a way that the part is most likely to function optimally until the end of the interval. But then – according to the law of statistics - the end of the usable service life has not yet been reached for many individual parts. With predictive maintenance, components can be used right up to the end of their service life – and replaced shortly before that. This extends the useful life, the parts need to be replaced less frequently, thus reducing operating costs and, finally, the environment is pleased when parts are used to their end. Predictive analytics and maintenance are complex digital tools, but they make a lot of sense and bring benefits, especially in highly critical production environments. For machine manufacturers, they offer the opportunity to offer customers tangible benefits with new business models.