Machine and system manufacturers usually have their reasons for not equipping the products they offer with modern cloud-based condition monitoring (condition monitoring) or at least with suitable data interfaces:

• The sensors and data technology required for this are too expensive.
• A machine or system would generate a lot of data on site, not all of which can be transferred to the cloud.

The need for additional sensor technology does indeed cause additional costs. However, the hype surrounding the Internet of Things in consumer electronics and other market segments means that suitable sensors are becoming increasingly affordable. In addition, most systems already provide a great deal of suitable data through the control system (PLC) that is used, although this has so far been hidden or isolated in the PLC.

The first step should be to gain access to this data and clarify in detail what information - relevant for condition monitoring - can be obtained from the existing data.

Example from practice

Figure 1: Some of the data required for cloud-based condition monitoring is already available in the PLC. The additional sensors required only incur minor additional costs.

© SSV Software Systems

A pneumatic subsystem for material transport in a production cell will serve as an example(see Fig. 1). It essentially consists of a guide cylinder with a pneumatic and PLC-controlled carriage that moves back and forth between the left and right end positions. At both end points of the guide cylinder there is a proximity sensor with a switching point to indicate the current end position of the carriage to the PLC (positions X1 and X4 in Fig. 1). LAN access via RFC1006 protocol (ISO-on-TCP) to the two PLC inputs for X1 and X4 is the only way to obtain the following information relevant for condition monitoring:

• Total distance covered by the carriage to date: The carriage on the guide cylinder has a maximum mileage, e.g. 3000 kilometers. By counting the end positions X1 and X4 reached, the total distance can be calculated and a statement on the possible remaining mileage can be derived.
• Exact number of all valve actuations in the valve terminal: For all valves belonging to the pneumatic subsystem, the number of valve switching operations can be counted using the end positions X1 and X4 and the possible remaining service life can be calculated according to the data sheet.
• Time interval for the slide movement from left to right and vice versa: The time measurements between the actuations of the switching contacts at the end positions X1 and X4 can be used to detect, for example, overpressure (slide too fast), underpressure or mechanical overload (slide too slow) as well as wear on the slide and guide cylinder.
• Shock absorber utilization: By calculating the carriage speed and counting the carriage movements between the end points X1 and X4, the drive-up speed and the maximum energy absorption per stroke and per hour can be roughly determined. In practice, however, this data is not sufficient to determine the remaining service life of a shock absorber.

Additional sensors required

All time measurements and calculations for the carriage speed are relatively inaccurate as long as only the digital proximity sensor signals of the end points X1 and X4 are available. The carriage travel time between these end points also includes the damping phases of the shock absorber stroke sections, which depend on various parameters. In this respect, the switching time of the proximity sensors at X1 and X4 is always delayed by the non-constant energy absorption time of the shock absorbers. However, the two shock absorbers at the carriage ends are also the critical components of the entire pneumatic subsystem. If the damping is no longer sufficient, the carriage moves unbraked to the stop of the guide cylinder. This can cause irreparable damage to the entire subsystem. It therefore makes sense to include the condition of the shock absorbers in the condition monitoring and to install additional sensors for this purpose.

If the effect of the shock absorbers is to be measured, the simple proximity sensors with one switching contact each should be replaced by a special variant with two switching contacts. During commissioning, the spatial distance between the two switches in a proximity sensor is set in a direct spatial relationship to the shock absorber stroke distance. This extension results in two new points on the X-axis, X2 and X3. The time spans (t1, t2) for each slide movement can now be determined to the millisecond. These stroke times will change with longer operating times and the associated reduction in shock absorber oil pressure and become smaller and smaller.

Pre-processing of the data

The status of individual assemblies and components can be displayed via a 'Condition Monitoring Dash Board'.

© SSV Software Systems

To enable predictive maintenance for the guide cylinder, all data should be transferred to a cloud service at certain intervals and stored there. Data pre-processing is helpful to counter the problem mentioned at the beginning with the large amounts of data. For example, suitable averaging for t1 and t2 over a certain period of time (e.g. two hours) can result in only two t1/t2 values being sent to the cloud service, even though a total of several thousand measured values were collected during this period.

Trend predictions based on evaluations of large volumes of data have been used in the IT environment for years under the collective term 'predictive analytics'. For this reason, there are highly developed and tried-and-tested services in various cloud service platforms that are also suitable for predicting the probability of failure of individual machine components and thus for determining suitable maintenance dates or as the basis for proactive service concepts. They fall into the Software-as-a-Service (SaaS) category and are available via various cloud and IoT service platforms.

In order to use predictive IT analysis services for predictive maintenance, suitable data must be collected or determined on site, transported to the cloud, time-stamped and stored in a database. The data does not even need to have a standardized structure for this. So-called NoSQL databases (e.g. Apache CouchDB, IBM Cloudant, MongoDB) are available as a service in the cloud. These document-oriented databases store unstructured data in JSON structures.

The trend prediction quality of a predictive maintenance solution depends firstly on the amount of historical data available in the cloud in order to generate a suitable prediction model using machine learning. On the other hand, data quality plays a major role. The more environmental data is available, the more precise the future prediction. For this reason, pressure, ambient and component temperatures - even the daily weather report - can also be included in the data acquisition for the pneumatic subsystem.

As you can see, the hidden digital values in the form of existing data must be recognized and used. Especially with regard to cloud-based condition monitoring and predictive maintenance, it would be a relatively small step to upgrade your own products with valuable new features in the form of intelligent data interfaces and services.