Condition monitoring has come a long way since the shop floor foreman listened to a machine via his screwdriver. In this article Marc Vissers of Lenze takes a more modern look at the subject.
The "old hands" on the shop floor can often hear before a breakdown if a machine is ailing or about to cause problems. But on today's automated factory floor there are smarter methods - and without any additional sensors too. Robots are increasingly conquering the shop floors in factories. They relieve employees of repetitive motion sequences or heavy loads. Or sometimes, they are simply much faster. So if the machine suddenly breaks down and production comes to a standstill, there are problems. One countermeasure consists of constantly monitoring the health of a machine or system, otherwise known as Condition Monitoring. Many machine builders find this method too complex and too expensive but with the latest technology this does not have to be the case, provided it is done properly.
Time and again, condition monitoring and predictive maintenance are used as synonymous terms, yet they are two different concepts. Predictive maintenance is the forecasting of events (or at least establishing the probability of events). Predictions make it possible to plan the replacement of the components in time, before the plant actually fails. Condition monitoring, on the other hand, is a preliminary stage. It is not a question of when things might fail. It is first and foremost about detecting - as early as possible - that the condition of the equipment is deteriorating.
Unfortunately, there is usually no way to directly measure the condition or "health" of a machine or a single component. Condition monitoring is therefore based on the interpretation of existing data in order to derive a meaningful description of the current condition. This requires a deep understanding of machines and processes in order to generate meaningful information from "bar" measured values. This knowledge is already available at OEMs who, on the one hand, know their machines and, on the other, understand the processes of the users. Analyses based on Machine Learning (ML) and Artificial Intelligence (AI) can help them to detect anomalies faster and potentially more accurately. The task now is to translate this know-how into an applicable and workable solution.
There are two different approaches. The first is model-based, centred on an assumed mathematical description of the machine, from which certain target values result that describe the normal state. The difference between these and the measured actual values allows a statement about the health condition of the machine. If certain tolerances are exceeded, this can be interpreted as a fault that triggers a warning in a condition monitoring dashboard.
The other approach is data-based. An algorithm first learns the behaviour of the system and the mutual influence of the measured parameters. In the case of a robot, these can be (for example), the speed, acceleration, torque, position or current consumption of a drive. The values measured during operation are compared with this learned description to define deviations. Of particular interest to OEMs is the fact that these approaches rely on the measured values already available and that no additional sensor technology is required to collect "health data". A drive that sounds different from the normal state does not need to be monitored with an additional sensor that detects the fault on the basis of a changed vibration profile. This is because deviations that indicate a worsened condition become visible as soon as the interaction of current consumption, acceleration and torque begins. So if it is possible to obtain the required information from existing data sources, the added value gained by condition monitoring does not have to be purchased at higher hardware costs.
How this data interpretation works in practice can be illustrated using a 2-axis robot that picks up loose workpieces and repositions them. Here, the stroke is actioned with a spindle drive and the transverse movement via a belt drive. A mathematical model can be sketched for the spindle drive, which describes (among other things) current consumption, torque and acceleration. These values should always remain the same for workpieces of the same weight. A possible defect, for example in the bearings, leads to higher friction on the spindle. In this case the measured values show a deviation, which is simulated in the showcase by a higher weight. The interpretation of the unusual actual values thus leads to the conclusion that there is a problem with the spindle drive and the user must be informed about it. On the second axis, an ageing toothed belt can cause the belt tension to decrease. Due to the increased vibration this causes, the entire system is subjected to stress and, for example, the positioning accuracy is affected. Here too, the condition monitoring system, which has previously "learned" the relationships between drive speed, acceleration and position, would detect a deviation and sound the alarm.
The two condition monitoring approaches differ not only conceptually: the question of how the data is evaluated is also different. The model-based evaluation is usually performed in the control system, because no high computing power is required and all relevant data is already available there. For the data-based evaluation, on the other hand, ML and AI analyses can be considered, either in an edge server on the shop floor or as a cloud application. Here at Lenze, we provide a comprehensive automation portfolio which not only supplies drives and the appropriate FAST Toolbox programming environment, but also allows the controller to be created easily. Further, pre-tested algorithms are included for various applications, which allow the rapid development of condition monitoring solutions. In addition, the manufacturer supports machine manufacturers in jointly developing their benefit-enhancing business models for condition monitoring services. Sometimes it is only the goal of increasing OEE and reducing unplanned downtime that is clear, but not the concrete issues that condition monitoring can help to achieve. At Lenze, we have the experience to assist machine builders in unlocking the information treasures that can be gained from the interpretation of data based on their process know-how and machine knowledge. We believe it also gives the OEM complete freedom of choice when it comes to hardware. This includes a range of differently dimensioned PLCs for model-based Condition Monitoring.
Data-based evaluation can also be performed locally if, in our case, the flexible Cabinet Controller c750 is used. This hybrid computer, which combines both classic PLC functions and a Windows 10 platform makes an additional industrial PC in the control cabinet unnecessary, and offers sufficient computing power for simple ML and AI applications.
The Lenze IioT Gateway x500 opens the door to the cloud for applications in which very complex models need to be calculated, or where several machines need to be compared with each other. Here too, the user is free to make their own decision. If a platform is already in use, for example Microsoft Azure or Amazon Web Services, the machine data can be evaluated there. But in our case, users can also fall back on the X4 solutions from Lenze. The advantage of this turnkey option lies in the integration of extensive functions for OEMs and users around the machine. With this approach, the collected machine data is first logged in the cloud and stored in data clusters. They are available for a variety of evaluations - such as condition monitoring. The solution supports the user in visualising the evaluation in self-configured dashboards. These can also be optionally accessed on mobile devices to give operators even greater freedom. Live monitoring and alarms can be set up via the cloud too. In the showcase, the digital twin is represented by a machine simulation and the problematic component is highlighted in colour. Further web services include X4 Remote, which, as the name implies, enables remote monitoring and maintenance. A further feature is the cloud-based asset management, which simplifies the administration of machines, access to documentation or spare parts ordering.
For efficient condition monitoring of a robot application or any other machine, additional sensors - which would drive up costs - are not necessarily required. Instead, the machine's devices function as sensors, so to speak, and then it is important to be able to interpret this data correctly and to refine it into meaningful information. The basis is provided by the combination of OT, IT and methods of data analysis. The decisive factor is the OEM and their machine and process knowledge, which enables the user to become the data scientist for their machines. A provider like Lenze, which is active in all the fields involved - hardware, software, networking and cloud applications - can provide its partners with valuable support.
Learn more at www.lenze.com.