Applying Predictive Maintenance Technologies to Improve Industrial Manufacturing

AjayRaghavan_PARC_5x7 Applying Predictive Maintenance Technologies to Improve Industrial Manufacturing
Ajay Raghavan, Strategic Execution Director – Systems Science Lab, PARC, a Xerox Company

As the Internet of Things (IoT) takes hold across the industrial sector, every manufacturing company should consider new ways to deploy cutting-edge technologies across software, hardware, and networking platforms to gain a competitive advantage.

Factories in the automotive, semiconductor, chemicals and other fields have long relied on robots and automation machines to improve the productivity, quality and safety of manufacturing processes. However, when any of these machines degrade, fail or shut down unexpectedly, it has the potential to cause productivity or product quality issues and disrupt the entire production line, resulting in significant unplanned downtime, economic costs, production losses, and even work injuries. That’s why it’s so important to detect potential issues and faults in factory machines before any failures occur.

The most common way to maintain factory machines today involves schedule-based maintenance routines that are executed on regular timetables. However, this approach can become inefficient for large production lines because unnecessary maintenance often leads to increased downtime while still not preventing all failures. On the other hand, reactive fail-and-fix maintenance can cause even more expensive instances of unplanned downtime. Just one hour of unplanned downtime in a high-volume factory for automotive, semiconductor or petrochemical manufacturing can cause operational losses greater than $1 million.

Low-cost IoT sensing technologies and advanced analytics tools provide a new option for predictive maintenance to increase asset availability and productivity by identifying inefficiencies, reducing unnecessary maintenance and unplanned downtime. Artificial intelligence in conjunction with physics-based system models can be used to integrate the many disparate data points, sift through all the noise, reliably detect and diagnose incipient machine issues, and provide actionable steps for users to optimally utilize their valuable factory assets and keep production lines running smoothly.

A key goal of predictive maintenance is to detect faults in the early stages and allow proactive planned maintenance actions to be performed during off-peak production hours. In addition, parts with long lead times can be ordered ahead of the anticipated failure, thereby not only avoiding potentially long interruption of production but also providing relief for inventory management. In this way, companies can maximize productivity, avoid product quality issues and expensive unplanned downtime. The goal is to continuously evaluate all system components throughout their lifecycles to identify particular components for service or replacement before they lead to performance degradation or failure.

Researchers at PARC recently collaborated with Panasonic on a project for unsupervised fault detection that can effectively identify the faults in industrial robots using current signals. The solution suite for Panasonic combines embedded sensing, complex system models and artificial intelligence technologies that can predict adverse system conditions with high accuracy, negligible false alarm rates and near-zero missed detections.

One challenge in detecting faults in industrial robots involves the difficulty of obtaining enough labeled training data under both normal and abnormal health conditions. The solution developed for Panasonic involves adoption of unsupervised machine learning algorithms that can learn from labeled training data over time in conjunction with physics-based system models. For this project, experiments were performed on an industrial robot under both normal and abnormal conditions. The results validated that the proposed fault detection framework was effective in detecting gear-wear faults in robots with higher than 96% accuracy.

In another example of condition-based maintenance, PARC partnered with Leoni, a global provider of energy and data management solutions for the automotive sector and other industries. Leoni engaged with PARC to explore predictive maintenance technologies – including system analysis, artificial intelligence and embedded sensors – to drive its digital transformation forward.

The vision is enabling Leoni to support its customers with actionable data and design optimization, as well as the ability to smartly monitor and manage their systems. This combination of new IoT technologies will allow Leoni to make energy and data transmissions in its industrial cable systems even more intelligent, efficient and reliable, with high predictive accuracy and negligible false alarm rates.

As part of the digital transformation, Leoni is leading its clients and partners in embedding intelligence throughout their ecosystems, including intelligent cables, cable systems and components for the manufacturing, automotive, energy and infrastructure sectors.

These two examples at Panasonic and Leoni illustrate how the industrial world is undergoing a rapid transformation based on the combination of low-cost sensors and advanced analytics engines. The purpose of predictive maintenance is to gain a better handle on the performance of critical devices and parts before something goes wrong on the factory floor. By combining smart IoT sensing systems with AI, manufacturers can reduce costly downtime while effectively increasing factory production.