The wealth of data from various sources in the manufacturing environment holds significant value, and with AI analysis, the benefits of these data streams can be quickly realized. The key is understanding the data to be correlated, and with this, choosing which information to reveal and when.
The confluence of machine learning and artificial intelligence, leveraged by the Internet of Things (IoT), and the vast volume of data that it generates, is creating a wealth of novel IT applications and opportunities. Nowhere is this more apparent than in the world of manufacturing, where this combination of advances has profound implications for performance.
Within the production environment, modern, cutting-edge manufacturing execution systems (MES) already create a detailed data trail to support better manufacturing processes. Harvested from a mix of sensors, automation systems, and data producing platforms such as Enterprise Resource Planning (ERP), tools and quality management systems (QMS), an advanced MES is in a pivotal place to create a Common Data Model (CDM) that brings together a whole ecosystem of information.
Used to manage the production lifecycle, the MES helps create a solid seam of information throughout every manufacturing step in the production process, from order placement right through to final dispatch. This data stream can be harnessed to boost efficiency, productivity and quality compliance among many other functions.
However, with the growth of the IoT and an ever increasing array of diverse connected devices delivering a prodigious wealth of raw data, it is the smart management of that data stream that can unlock ever bigger gains in the production environment.
Additionally, with a CDM all actors of the shopfloor can join the MES in producing structured data. A CDM creates an open ecosystem, where other platforms and systems in the shopfloor can easily move from passive and siloed, into actively sharing their insights and information into the wider shopfloor.
The Power of Data
Properly configured, an IoT data platform has the capability to collect all types of data that are not only related to specific products and manufacturing process steps, but which can also include environmental data like humidity and ambient temperatures, as well as parameters associated with individual machines, such as oil temperature, power consumption, and vibration. It is easy to see how gathering, for example, energy consumption, cadence, and the temperature of a piece of equipment, together with the humidity of an area, yields the possibility of correlating all this data to derive actionable conclusions regarding the performance of that particular machine.
A big leverage of the MES is also the ability to go outside of a particular machine and to correlate against, flows, products, previous process machine and so on. Enabling different levels of analysis and meta-analysis. A simple use case would be a machine that receives materials from different equipment, the algorithms can discover that the information of what was the previous resource, or line is key to understanding why the machine has a higher probability of defects for some materials. The MES has context of everything that is happening in the shopfloor, therefore the context for inference is much wider than algorithms dedicated to a particular machine or vendor can achieve.
Collecting historical data can thus enable a machine learning algorithm to correlate this data to learn about specific outcomes and ultimately make predictions about the future. For quality control, like trends can be detected, and the algorithm can predict that in, say, 20 or 30 products, some parameter of the resource will be in non-conformance. This moves to improve on other statistical based approaches like Statistical Process Control (SPC) that are very narrow in their context of analysis.
An appropriate IoT data platform, therefore, has the capability to collect data from multiple sources, including the quality control aspects of the MES, and then correlate this data with external parameters such as the vibration of a specific bearing, to support informed quality control decisions. The machine learning algorithm may determine, for instance, that quality declines as vibration goes up and the increase in vibration is associated with higher temperatures. With this conclusion in place, it is then possible to direct maintenance operations to examine, in this case, a specific bearing and either repair it or take action to keep it cooler. This type of analysis can be with hundred of parameters and millions of datapoints, the algorithm will be the one responsible for understanding and to surface patterns that affect the defect prediction.
The clear advantage with this approach is moving from reactive or scheduled maintenance and quality activities to predictive. Enabling predictive maintenance and predictive quality that can automatically create a maintenance activity request, for example, before a failure occurs that will halt a production line for a day, a week or even longer. There are other advantages of this approach, though. Consider a maintenance event that involves buying replacement parts and that will need to have a specific technician on-site and available to complete their installation and machine recommissioning.
By correlating all this additional data, the system will not only issue an alert with a certain level of confidence that a problem may occur and trigger an associated maintenance event, but it will also request the ordering of all necessary parts needed to replace or repair the equipment. It will also be able to ensure that when those parts are available, the relevant maintenance workers are on hand to promptly execute the maintenance activity. The equipment is therefore stopped for the minimum amount of time possible. Such measures improve the efficiency and availability of individual pieces of equipment and inevitably lead to significant improvements in the overall quality and productivity of a manufacturing environment.
Another use case of the bulk IoT data may emerge following a customer complaint or a product return. Executing a program of measurements and analysis across the shop floor can help to determine the cause of a problem with a particular delivered product. However, by interrogating the data on the various factors that influenced the outcome, the data can be correlated to try to reveal all the other products that have the same characteristics and the same types of values. This may enable a product recall to be announced before multiple complaints emerge and can thus help to quickly manage problems when they arise. This approach gives better control over reputation management, for example.
Layered Data
Given that sophisticated IoT platforms should be able to harness the power of data and AI to avoid quality problems by creating algorithms to improve maintenance activities, boost performance and address issues in a timely way, the value of this data becomes clear. However, it is also evidently unwieldy to attach that vast bulk of operations data to an individual product within the MES quality control system. While this data is undoubtedly valuable and cannot be discarded, MES operators typically only require traceability history, work instructions, and quality control. Detailed maintenance logs for each machine, such as, power consumption, temperature, or humidity levels are not relevant within a product’s electronic history record.
An appropriately functional MES will, therefore, present two data sets. One set of data is associated with a specific product, such as a Configuration to Order (CTO) product, and contains all the quality control and associated manufacturing data that needs to go with that product as it runs through the production process and through to final dispatch.
Beneath that runs a large volume of other operational data that doesn’t necessarily need to be associated with a specific manufactured product. However, this data provides essential context for quality control by contributing to a broader analysis of maintenance and system-wide performance. This information helps to reduce errors, rebounds, and the need for remanufacturing. All the available data helps to confirm that products are manufactured with the set performance limits and have a significant role to play in maintaining that performance baseline. While this information is highly relevant to the MES, it doesn’t always need to be visible to operators. The key then is to derive value from data but to rationalise where and when this data is displayed and carried.
Navigating the Data Lake
Across the shop floor, there is a clear trend for the industry to gather steadily more data from an even greater number of sources. Using machine learning and AI, it is possible to take advantage of all the data that is being collected to detect trends and the impact different parameters have on quality, speed, and the availability of equipment. This data can be leveraged to predict the future and avoid any potential issues before they occur.
An advanced MES manages to align all of this data in a single compatible format to not only avoid so called data silos but also make all the data available, actionable and valuable. Coupled with AI, the outcome is a system of data-led decision making system that utilizes machine learning to support conclusions from a vast lake of data points without overwhelming the operators by excessive details. With mass data increasingly available in prodigious volumes, it’s all got value, but only if it can be applied in the right way.
