Machines are becoming increasingly intelligent - but the occurrence of a malfunction can never be completely ruled out. If this happens, it is always costly and time-consuming; however, it is all the more annoying if the necessary machine data and production parameters are missing to analyze and, in particular, prevent the error in the future. The consequence is often that the behavior can no longer be analyzed and, as a first step, only additional data logging mechanisms are installed. As a result, the misbehavior must occur again in order to continue the analysis.

This information deficit can be avoided by recording all process-relevant data of the machine with cycle accuracy - as is the case with the 'Twincat Analytics' solution from Beckhoff.

No analysis without data

Twincat Analytics offers numerous application scenarios: Storing and analyzing data directly on the local controller, in private networks or in the public cloud.

© Beckhoff Automation

This creates a complete record of the processes in the machine. Depending on requirements, the data can be collected and analyzed locally on the machine computer, within a cloud-based solution in your own network or on the internet. The cloud-based approach is particularly suitable for developing new business models. This is because it is not only possible to analyze behavior retrospectively, but also to analyze the data in order to take preventive action on the machine in question. In other words, the keyword is predictive maintenance, which machine manufacturers can offer their end customers as a service.

The basis of an efficient analysis is complete data. To obtain this data, in the example of the Twincat Analytics solution mentioned above, the so-called 'Analytics Logger' is activated on the control computer after it has been configured in the Twincat 3 engineering environment. In the configuration interface, the user simply uses a checkbox to specify which data from the process image or the application should be recorded cyclically. The user can also specify whether the data should be saved locally or sent via a communication protocol. A ring memory can be set for both applications. This is useful for local storage in order to comply with the maximum possible storage capacity. If the data is communicated directly, the ring buffer can bridge a connection failure accordingly.

This creates a complete record of the processes in the machine. Depending on requirements, the data can be collected and analyzed locally on the machine computer, within a cloud-based solution in your own network or on the internet. The cloud-based approach is particularly suitable for developing new business models. This is because it is not only possible to analyze behavior retrospectively, but also to analyze the data in order to take preventive action on the machine in question. In other words, the keyword is predictive maintenance, which machine manufacturers can offer their end customers as a service.

The basis of an efficient analysis is complete data. To obtain this data, in the example of the Twincat Analytics solution mentioned above, the so-called 'Analytics Logger' is activated on the control computer after it has been configured in the Twincat 3 engineering environment. In the configuration interface, the user simply uses a checkbox to specify which data from the process image or the application should be recorded cyclically. The user can also specify whether the data should be saved locally or sent via a communication protocol. A ring memory can be set for both applications. This is useful for local storage in order to comply with the maximum possible storage capacity. If the data is communicated directly, the ring buffer can bridge a connection failure accordingly.

Direct data transmission via the analytics logger is particularly suitable for the development of new business models. It uses IoT communication protocols, which offer ideal properties for the use of cloud services. The IoT protocols always establish an outgoing connection to a message broker. This decouples the communication so that, unlike classic client/server communication protocols, the participants do not need to know each other. The communication participants all act as clients. In this case, the analytics logger on a control computer is an IoT client that 'publishes' data to a message broker and stores it in a topic. Topics can be structured hierarchically. An example of such a topic: myCloud/CustomerA/WoodWorkingMachine9/PackagingModuleB/Data

The message broker itself only holds a list of 'interested parties' for the corresponding topics, as other IoT clients can 'subscribe' to these topics or their data. For example, an analysis server may be interested in the logger data or a mobile application on a smartphone. Both are IoT clients, subscribe to a corresponding topic and each receive a copy of the data.

The practical thing about IoT protocols is the outgoing connections, as only incoming connections are usually blocked by firewalls. This eliminates the need for time-consuming activation of ports. Another resulting advantage is the flexibility that can be achieved with this technology: The same mechanisms can be used within your own local network architecture as well as for communication with services on the internet. Cloud providers such as 'Amazon Web Services' or 'Microsoft Azure' have their own IoT message brokers that can be used for communication. The best-known protocols are currently MQTT (MQ Telemitry Transport) and AMQP (Advanced Message Queueing Protocol) - both are supported by Twincat.

The analytics infrastructure

Significant points are marked in the charting tool to make them easy to find.

© Beckhoff Automation

As already mentioned, the IoT interface gives machine manufacturers and end customers a great deal of freedom when setting up an analytics solution. The recorded data can be analyzed locally on each machine with the corresponding PLC library. If some machine control systems are not powerful enough for local analysis, the data can be analyzed directly at the end customer's site via the IoT connection in a local cloud. This allows machine operators to analyze their machines themselves in their own network environment. In this case, the software can run on a server and carry out the analysis of several machines at this production site.

Alternatively, the solution can be installed on a virtual machine. This also makes it possible to use a public cloud. Processor power and memory as well as the IT infrastructure can be flexibly rented and used from providers such as Microsoft Azure. This significantly simplifies the global connection of machines to the analysis system. Another option is for a machine manufacturer to act as a service provider for its machines and analyze the machine data in the cloud or use the cloud only as a 'transmission medium' to perform the analysis on a server in its own IT infrastructure. If an end customer - who is interested in high machine availability, productivity and product quality - prefers to employ an external analyst, they can also disclose the access data to the message broker, the topics architecture and the data description. A 3rd party analyst is thus able to obtain the necessary data and offer his customer the corresponding services.

Big data needs to be mastered!

The Analytics Workbench is based on a runtime that can be configured and programmed using the Twincat engineering environment. The great advantage of this is that machine builders do not have to change when switching between the programming environment of the machine control system and the environment of the analytics software. This means that the programming know-how built up over the years can be applied 1:1 in the Workbench. This also makes it very easy to write your own algorithms for the analysis or to reuse algorithms already used in machines. Alternatively, you can use the algorithms in the Analytics PLC library. These include modules for counting edges, analyzing minima and maxima, assessing machine cycles over time or calculating the energy consumption per time of a selected component. Information such as the shortest, longest and average running time is particularly helpful when evaluating machine cycles over time. This allows optimization potential to be derived directly or indicators for predictive maintenance to be identified. For example, whether a milling head is frequently stationary, tends to run at speed a, b or c or is often in an error state can be determined very easily using a condition analysis. The results can be clearly displayed in a histogram, which is why the 'Scope' charting tool familiar from Twincat also plays a key role in the Anayltics Workbench. This applies in particular in conjunction with the Analytics Configurator, which is also embedded in the engineering environment. With the configurator, it is possible to put together a post-scope configuration based on data that has already been recorded in order to be able to display the data curves graphically again.

To 'view' the data, the Analytics Configurator uses the same algorithms as those used in the Analytics library. The selected time ranges of the data streams are analyzed and displayed directly in the configurator. Significant values identified in this way can be dragged and dropped into the charting interface of the scope. The scope then automatically jumps to the relevant points to graphically illustrate the context to other signals. This makes it much easier to find the needle in the 'big data pile'. In addition, engineering is significantly simplified with the Analytics Configurator. As all algorithms originate from the same source, it is possible to import the configuration set in the configurator with all selected variables and the associated limit values into the PLC in order to switch from an offline analysis to an online analysis with streamed data from the cloud.

The functionalities described refer to the Analytics Workbench Base. It includes a PLC runtime, the Analytics PLC library, an IoT connection for streaming data, the Analytics Configurator and the 'Scope View Professional'. The workbench can also be expanded with packages for condition monitoring, C++ and Matlab/Simulink. In particular, the Matlab/Simulink integration in the runtime allows very extensive access to useful toolboxes on the subject of analytics. For example, there is a toolbox for machine learning or optimization. In addition to the special extensions for analytics, other standard Twincat on-board tools can also be used. The database server is able to store data online and offline in various databases. In addition, an analytics system can be 'fed' with data via OPC UA, which is widely used in automation, and there are also converters from OPC UA to IoT protocols to enable third-party controllers to access the analysis, for example. Last but not least, data visualization is a very important point: Twincat 3 HMI enables the user to design intuitive dashboards for the Analytics Workbench based on HTML5. This ultimately results in an analysis cockpit that can be used to display all results for a machine or across machines. A hierarchical structure then allows the presentation of ever more precise details. Ultimately, machine data and its analysis are the key to many new business models and future-oriented, more efficient automation.

Author:
Pascal Dresselhaus is Product Manager Twincat at Beckhoff.