Industry 4.0 and the Industrial Internet of Things (IIoT) promise to make more data and better analytic applications available for industry. Driven by high-powered analytics, including machine learning and artificial intelligence engines, both promise operational benefits, including lower costs through improved key performance indicators (KPIs) such as first pass yield, non-standard downtime or overall equipment effectiveness (OEE).
However, despite the investments in data capture and analytics, the industry faces a daunting challenge in attempting to drive continuous improvement, given the amount, types, and lack of context in most manufacturing data. The result on the shop floor is often characterized by data overload and inaction.
Key questions for continuous improvement
Plant operations staff need to confidently answer such challenging questions as:
- How do I know my process is in control?
- How do I sort through massive amounts of data to find root causes?
Answers to these lead to actions not only to improve the KPIs of current manufacturing operations, but also to the development of more robust designs for future products. If there are multiple lines or plants, making products the same way across various work centers is an additional essential objective.
In a paper Data Analytics Gives Manufacturing Plants Insights for Continuous Improvement, we present a proven strategy for continuous improvement in manufacturing that can provide:
- Insights into process variability
- Correlations with product quality measures
- Real-time statistical process control (SPC)
- Hierarchically structured data for regulatory compliance
- Powerful machine learning algorithms to handle massive and disparate data common in process industries
Incorporation of this strategy into continuous improvement systems (Lean, Six Sigma, etc.) has led to bottom-line improvements in hundreds of work centers in a global infrastructure.
An example from a chemical reactor is presented to illustrate the methodology and application in both an on-premise and cloud-based environment. Techniques using both univariate and machine learning algorithms illustrate how to explore and monitor complex manufacturing data sets.
Key learnings
Data analytics including machine learning algorithms can deliver breakthrough improvements by making the data exploration phase efficient and real-time monitoring and prediction easy. As the IoT multiplies the volume and type of data available to process manufacturers,
these analytic algorithms provide capability to continuously improve and optimize manufacturing operations. Inspection of univariate control charts exposes underlying process variability.
Coupled with event synchronized drill downs to the time domain, operations teams can deploy real-time SPC. Machine Learning techniques offer powerful algorithms that can generate insights into process variability and correlation to quality measures. These can lead to enhanced understanding of process variability, predictions of process failures, product correlations, and insights to improve future products with reduced commercialization times and more robustness in full-scale production. Download the white paper on this topic from Optimation’s website.
This article was provided on behalf of Optimation, a certified member of the Control System Integrators Association (CSIA). The author, John Scheible, is principal engineer for MVE Analytics. He has worked in the chemical industry for more than three decades in design engineering, process engineering and commercialization of specialty chemicals at Eastman Kodak. For more information about Optimation, visit its profile on the Industrial Automation Exchange.
