Today, production maintenance is essentially based on two approaches: reactive and preventive maintenance. However, both methods have disadvantages, particularly in terms of possible downtimes and costs. With reactive maintenance, components are operated until they fail and only then are they replaced. This means there is a high risk of unplanned downtime, which can result in very long downtimes - for example, because the specialist responsible is not available, troubleshooting is more complex than expected or the spare part is not in stock. This type of maintenance can result in high costs in an interlinked process in which the failure of one process step leads to the shutdown of the entire production process.

In preventive maintenance, wearing parts are replaced in fixed cycles to prevent unplanned downtime. If the service life of the components is known, this method can prevent unplanned downtime with a high degree of certainty. However, this results in high maintenance costs, as parts are replaced that might have worked for a long time. The method is also time-consuming and can result in longer, albeit plannable, maintenance downtimes. The biggest difficulty is defining the correct maintenance interval. If it is too long, breakdowns will occur. If it is too short, the maintenance costs are high. This means that a great deal of experience is required to correctly estimate the service life of wearing parts.

Sensor measures wear

It is more efficient if users can recognize the optimal replacement time in good time and plan the time of replacement in production. The 'magic word' is predictive maintenance. This involves continuously monitoring components that are critical to failure and determining a deterioration in their condition based on characteristic wear features.

Wear monitoring on motors in machines is already well established: if the motor bearings wear, the vibrations increase. These can be measured using a vibration sensor and the data sent to an evaluation system, where mathematical algorithms are used to calculate the best time to replace the bearing. In addition to motors, however, there are many other parts in machines and systems that are subject to mechanical wear. These include, for example, moving electrical cables in drag chains, which supply moving machine parts with power or data. During maintenance, these must be replaced regularly to avoid unplanned production downtime. The replacement cycles are primarily dependent on the frequency of movement, temperature, bending radius, travel speed and acceleration. A suitable cable design and the use of high-quality materials also have a significant influence on the number of possible bending cycles.

Predictive maintenance for cables

The 'Etherline Torsion Cat. 7' is a high-speed cable for industrial Ethernet.

© Lapp

Data cables such as Ethernet cables are considered to be particularly prone to failure due to their complex structure and the necessary high-frequency properties. There are many reasons for failure: broken shielding, for example, leads to increased EMC interference. Broken wires initially lead to increased attenuation and reduced data rate and later to total communication failure in the event of complete wire breakage.

In addition, predicting the durability of cables in real applications is very imprecise due to the diversity of applications. If a service life of five million bending cycles is specified for a data cable, this is done by specifying certain typical values for the bending radius, acceleration and travel distance parameters. However, if the cable is operated differently - for example at a significantly lower temperature or with a longer travel distance - a lower number of bending cycles can be expected. If, on the other hand, the conditions in the application are less demanding - for example due to shorter travel distances and lower accelerations - more than five million cycles can also be achieved. Deviations from several parameters can also lead to interactions, the effect of which cannot be predicted. It therefore makes sense to include cables subject to wear in a predictive maintenance system.

No change to the line

Lapp has now developed a process for the predictive maintenance of Ethernet cables subject to wear. Although the focus is primarily on cables in drag chains or robots, the solution can also be used to monitor other critical cables where failure is to be avoided.

The first development goal was to develop a measuring principle that works without changing the cable, i.e. without additional measuring wires in the cable. This is advantageous in that standard Ethernet cables and standard connectors such as RJ45 or M12 plugs can be used. The installer connects the cables as usual and does not have to connect any additional test leads. This is because one difficulty with such test leads is establishing a correlation between the wear of the test leads and the useful leads. There is a risk that the useful cores for the actual data transmission will fail earlier than the measuring cores, resulting in an unplanned standstill. It was therefore clear from the start of the project that Lapp's system should measure the actual transmission characteristics of the user wires. This also makes it possible to retrofit existing systems.

Another challenge is that the measurement must be carried out during operation so as not to disrupt system operation. For rough monitoring of the transmission channel, current methods in managed Ethernet switches from Lapp can already measure the bit error rate. The proportion of erroneous bits per time unit is determined. However, this only allows a very imprecise prediction of the line status - when bit errors occur, the line damage is often already very advanced and the necessary lead time for maintenance planning is very short or imprecise.

Box analyzes data packets

The system from Lapp measures up to four transmission-relevant parameters. Thanks to the measurement of several variables, plausibility checks are also possible so that misinterpretations of measured values can be minimized.

The measurement process is integrated in the so-called 'PMBX' (Predictive Monitoring Box). It has two Ethernet ports and is simply inserted at the start of the Ethernet cable to be monitored. The data packets are transmitted transparently from one Ethernet port to the other in cut-through mode, i.e. with virtually no delay. The 'PMBX' is not visible to a connected PLC and has no influence on the data transmission. This means that it can also be integrated into existing systems without requiring changes to the PLC software.

In the next step, the measurement data is analyzed. The predictive maintenance system uses a deep learning approach for this. For Lapp's drag chain cables, millions of data records are collected in the in-house test center and then analyzed using mathematical algorithms. The data is analyzed locally on a PC during the development process, but can also be run in the cloud later, depending on customer requirements. The more data is available, the more accurate the prediction becomes. The system is self-learning. After collecting data for just a few weeks in the test center, it was possible to predict cable failures with an accuracy ranging from a few hours to several days.

The solution can therefore help to reduce machine and system downtimes. The system is currently being further developed to meet the requirements of various applications. It is being tested in practice together with pilot users and the prediction period is being optimized by collecting further data.

What are the benefits of predictive maintenance?

Predictive maintenance process for data cables from Lapp.

© Lapp

• Maintenance can be planned. Maintenance is carried out when production is not taking place, for example on Sundays or evenings, or when a shutdown is planned for other reasons anyway.
• Necessary materials do not have to be kept in stock and can be ordered just in time when they are needed.
• The costs for spare parts are reduced because components can be operated until the 'end of life' and no safety margin needs to be planned for.
• The risk of failure is reduced and the productivity of systems is increased because the actual condition is known.

Author:
Ralf Moebus is Product Management Automation at Lapp in Stuttgart.