Digital Transformation of Legacy Assets in Manufacturing
By John Traynor, VP & GM, Industrial Solutions, TensorIoT and Philip Weber, VP, Alliances, CloudRail
One of many challenges facing manufacturing firms exploring “Industry 4.0” and “smart factory” digital transformations is the variety and vintage of equipment on the production floor. Industrial assets generally have very long capital replacement cycles measured in decades, whereas technology advances at a pace measured in months and years. Despite the challenges, the benefits to bridging the OT/IT gap include broader observability of assets, increased situational awareness, and deeper insight into maintaining and improving operational performance. In this blog post we discuss how you can quickly and easily ingest data from production equipment and put it to good use to improve operations.
There are three common data ingestion situations. Let’s consider each of them and then discuss the preferred approach.
One, if your factory is shiny and new, you may be able to get all your data from a modern protocol such as OPC UA. While this works for newer assets, most equipment on the production floor today is older and does not support this approach.
Two, it may be possible to read data directly from a PLC (programmable logic controller) or DCS (distributed control system) embedded in the equipment or controlling the integration of multiple pieces of equipment. This can be done either by using an existing data protocol (such as Modbus TCP, Profibus PA, FOUNDATION fieldbus, and others) to read data as it moves between equipment and controller, by adding hardware to read directly from PLC registers, or by modifying PLC programming to deliver the desired data to an external resource. Often this approach is fraught with unforeseen complications which waste time, require highly skilled automation technicians and engineers to address, and risk interfering with existing operations if not handled correctly.
Three, and typically the most straightforward approach, is to add secondary sensors to equipment. This is especially useful for the oldest assets which don’t have a modern data interface, including older machines which are purely pneumatic or mechanical, as well as newer but simpler tools used in manual operations, such as manual tube bending equipment and steel-cutting chop saws. There are thousands of sensors available, with the most common including sensors to detect electric current, proximity, distance, vibration, pressure, and temperature. Variances in sensor data can be used to infer normal or abnormal performance.
Of the three methods outlined above, the first and third are the least expensive and easiest to implement. The second method is time-consuming and labor-intensive and should be used only as a last resort.
Regardless of the method chosen to ingest data, it is often tempting to collect as much data as possible. In practice, significant gains can be made with just one or two data points. If more data is needed, it is often more economical to go back and add an additional data tag or sensor later, rather than wasting resources to capture data that will never be used.
In addition to selecting the data and method by which it is ingested, it is also important to consider how frequently data needs to be collected. Getting a total part count per shift may not provide enough detail to understand how production varies during the shift, so a measurement once per minute may be more useful. Updating that same part count every 100 milliseconds will not, in most cases, provide any additional insight.
Once a data ingestion path and frequency are selected, the next step is to give each data point meaning via an asset model. For example, the number 72 could be the internal temperature of a pump, the speed of a conveyor, or the humidity of a paint booth. An asset model is critical to associate a measurement and the measurement’s units with the right piece of equipment or phase in a process.
Storing and Processing Data
Next, it’s time to decide how to store, process, and use the data. The use of scalable, serverless, pay-as-you-go services in the cloud is the simplest approach. As a leading cloud provider, AWS offers several cloud-based services to ingest, store, and process IoT data without requiring you to setup or provision underlying infrastructure. For example, AWS IoT SiteWise provides storage of data, asset modeling, and automatic aggregations and calculations on incoming data. This pay-as-you-go approach allows pilots and proof-of-concept projects to be run at low cost.
The easiest approach to reliably moving data from the production floor so it’s available for visualization and analysis is to purchase a qualified device bundled with the required software.
CloudRail provides a packaged approach which can easily handle data from OPC UA servers as well as from IO-Link sensors. Provisioning is done via the CloudRail device management portal, and data is delivered directly to your own AWS cloud account, so you are always in control of your data. With a few clicks, data can be sent automatically to a variety of AWS services, including AWS IoT SiteWise.
There are thousands of sensors available to measure everything from temperature, pressure, distance, vibration, flow rate, and more. As mentioned before, a lot of insight can be gathered with just one or two sensors.
TensorIoT and CloudRail recently worked with an aerospace company which created parts using CNC machines. The company lacked insight into machine utilization and its true manufacturing capacity. By adding a simple vibration sensor to each machine the company can now understand when machines are active and see exactly how much idle capacity is available across all machines.
Gaining Insight from Data
Once you have data landing in the cloud, you can build your own application or use an off-the-shelf solution deployed in your own cloud environment. For example, TensorIoT provides its SmartInsights solution to visualize and generate insights from raw data and provide a global view with multiple sites, assets within each site, and properties associated with each asset.
Often just seeing live data — previously hidden inside the machine — provides an initial “a ha” moment of insight. We recently worked with a client who immediately recognized that certain equipment was not configured for optimal use, but previously lacked the insight to take action.
After the most obvious opportunities for improvement have been addressed, more advanced methods, such as the use of machine learning, can be added to the system. There may be a large number of confounding factors which make it difficult to do root cause analysis or spot variations in machine performance by observation alone. These more challenging circumstances can be ideal candidates for using machine learning for anomaly detection.
Conclusion and Next Steps
Industrial IoT and Industry 4.0 hold great promise for helping companies improve performance. There is so much opportunity it can be sometimes difficult to know where to start. One key piece of advice is to avoid the temptation to “boil the ocean”. In practical terms, that may mean connecting just a single machine, or a small cluster of related machines. We call this the crawl, walk, run approach. An advantage of this approach is that you can start to see the benefits of data collection and visualization immediately and with minimal investment. The initial benefits realized can then provide the economic incentive to expand to more equipment and add additional capabilities.
For guidance on how to start your Industrial IoT and Industry 4.0 journey by gathering data from your existing machines and equipment, regardless of age or type, reach out to us at email@example.com.