Actionable Insights Reduce Downtime



How machine learning can discover patterns hidden in time series data to unlock operational improvements

The next industrial revolution is here. Whether you call it Industry 4.0 or Industrial IoT or Digital Transformation, the increased access to machine and operational data, proliferation of two-way communication, and speed of data flow, combined with the lower cost of computing, connectivity, and storage have created the perfect environment to transform industrial operations. The time series data generated by these operations, if harnessed effectively, can provide actionable insights to reduce downtime as well as improve throughput, operator safety, and product quality.

Finding Patterns in Time Series Data
The time series data generated in discrete and process manufacturing operations is very rich in information that can provide insights on the current and future health of the production equipment and lines. Nearly all operational systems (e.g. manufacturing lines, industrial equipment, data centers, handheld devices) produce streams or bursts of time series data in the form of sensor readings, log entries, or activity traces.

However, much of this data goes underutilized as traditional approaches such as regression models, statistical process control (SPC), and optimization have limitations to effectively leverage multivariate trends and uncover new insights for improving operations.

A key source of insights in time series data comes from multivariate trends, also called patterns, which are often reviewed forensically to understand past system behavior. Patterns are often hard to describe and need to be learned and correlated across many signals in the data. As a result, they are not identified and used by traditional analytics approaches.

Pattern Discovery—The Cornerstone of Predictive Analytics
Overcoming limitations of traditional analytics, machine learning makes it possible to discover new patterns hidden in time series data, including those that would go undetected by the human eye, correlating them to operational events.

Pattern discovery plays a critical role, particularly for assets, in identifying patterns prior to degradation occurring. This is crucial as it moves asset management from reactionary (degradation has occurred, and you attempt to minimize its impact) to proactive (you prevent it from occurring).

Figure 1: Machine learning enabled pattern discovery and early warning

By identifying patterns that are precursors to undesired events or operating conditions, these predictive systems can provide actionable insights on the current and future health of the production systems and the products they create. With this is in place, an operational machine learning system can perform pattern discovery, condition monitoring, and predictive analytics in real time on existing operational data.

Putting Machine Learning in the Hands of Industrial SMEs
Most commercial machine learning solutions today are delivered as platforms or require you to have data scientists on staff to prep the data, model the system, interpret results, and make operational recommendations. This approach poses two challenges that lengthen the time needed to see value.

  • Platforms get in the way of existing workflows or other digital initiatives and can disrupt critical operations, as they require more resources, proprietary instrumentation, and time to deploy.
  • Data scientists are hard to come by (and expensive) and, even when available, are often not domain experts in manufacturing operations.

Industrial subject matter experts (SMEs), on the other hand, understand and often design operational processes and best practices. These high-value manufacturing, operations and process engineers have specific knowledge of how to operate equipment, execute maintenance, and ensure safety. However,  industrial SMEs are often overloaded.[1]

Solving the challenges noted above can happen when industrial SMEs are ready to use the machine learning system without much effort. The SMEs can feed into the machine learning system existing multivariate time series data generated by their operations. Then, based on the system’s predictive insights and alerts, SMEs can determine the corrective action to be taken. Such an approach is effective for three reasons:

  1. The industrial SMEs retain both the control and visibility needed to drive improvements.
  2. The SMEs are in the best position to determine the corrective action to take.
  3. A shorter time to value is realized, as the entire process, from identifying the use case to verifying the results, resides within the operations team.

Use Cases Across Industries
Since operational machine learning discovers patterns in available time series data, it does not require developing mathematical models of a physical system or demand that vendors possess deep domain expertise in a particular industry. As a result, it is highly versatile and can be applied across many industries and use cases.

Figure 2: Use cases across industries

Quality Improvement
Challenged by variability in quality and yield, a large chemical manufacturer believed the high volumes of process and machine data collected for different phases in its manufacturing process should help identify problematic batches early. However, traditional efforts had proved ineffective.

Integrating the customer’s process historian to the operational machine learning system enabled rapid analysis of thousands of batch executions. This helped identify hidden patterns and early warning signals for each phase of the manufacturing process. By providing predictive final quality results early in the manufacturing process, the customer was able to scrap low-quality batches early, increasing process efficiency, quality, and yield while avoiding downstream supply chain issues.

Throughput Improvement
A mining company was experiencing frequent, unplanned downtime due to variation in raw materials that impacted a critical process-line machine. This downtime was costing the company $30,000 per hour. Instrumentation and data collection captured large volumes of operational data, but efforts to turn this into meaningful improvements in operational efficiency fell short.

The operational machine learning system was integrated with the customer’s data historian. Analyzing and correlating the data streams representing motor currents, temperatures, and valve settings with downtime events enabled the customer to identify patterns leading to the downtime event. Using this condition monitoring, the operations team took corrective actions, improving machine uptime and throughput.

Predictive Maintenance and OEE
A semiconductor manufacturer was operating complex, expensive equipment that executed many different types of step-based operations daily, and optimizing utilization was a high priority. The machines were instrumented to collect operational data every second in the form of sensor readings, control parameters, and other settings.

The operational machine learning system was easily integrated with the manufacturer’s operational data store containing trace data, quality measures, and inspector and operator log information. A four-month history of data created a model identifying multiple abnormal conditions known to create maintenance indicators, creating alerts in the Falkonry system. The availability of advance warning provided by the Falkonry assessment stream supported early intervention by the maintenance team, improving uptime and Overall Equipment Effectiveness (OEE).

Conclusion
While dealing with multivariate time series data may seem complicated, machine learning addresses much of that complexity. And now with “ready to use” operational machine learning, historical data can be selected, patterns discovered, and models built and verified—all without the need for a data scientist or third-party consultants, which significantly reduces the cost. In addition, putting machine learning in the hands of SMEs delivers a shorter time to value since the entire process, from identifying the use case to verifying the results, resides within the operations team. This approach is delivering substantial improvements in asset performance, throughput, operator safety and product quality.

[1] https://industrial-iot.com/2017/12/industrial-subject-matter-experts-and-analytics/


Sanket Amberkar is SVP, Marketing at Falkonry. He leads marketing at Falkonry and is responsible for the company’s positioning, thought leadership and go to market strategy.

Amberkar has over 20 years of experience in the high tech, energy, industrial, and automotive markets in areas ranging from product development to market strategy. Prior to Falkonry, he was VP of Product Marketing for Innovation & New Ventures at Flex, where he brought to market its Innovation services and launched the LabIX startup initiative. Earlier he led marketing and product development teams at Cisco and Delphi.

Amberkar holds Master’s degrees in Electrical Engineering and Business Administration—both from the University of Michigan. He is a frequent industry speaker and holds thirteen U.S. patents.

Share and Enjoy:
  • Digg
  • Sphinn
  • del.icio.us
  • Facebook
  • Mixx
  • Google