It sometimes seems that the exuberance for the Smart Factory has overshadowed the reality on the factory floor. If you are to believe the analysts, we will witness a surge in industrial technology spending in the next few years, which will be fueled by digitalization and Industrial IoT. McKinsey estimates that by 2025, digital transformation will have freed up between $1.2 – $3.7 trillion in global productivity gains for the manufacturing sector.
Regardless of whether these projections are accurate, the fourth industrial revolution is underway and plant owners will need to adjust to the new reality to remain competitive.
What does this mean for asset maintenance?
The hybrid approach to asset maintenance
In the past, we have addressed the topic of the Hybrid Smart Factory and why it is prudent for factory owners to take an incremental approach to investing in Industrial IoT and digitalization. Even as we enter the age of machine learning, there are two major inhibiting factors:
Skilled staff. At the core of machine learning are large amounts of data. Industrial plants are struggling to recruit big data scientists and engineers who typically have more lucrative opportunities.
Cost. Technology giants are spending billions of dollars on R&D in the expectation that tomorrow’s factory owner will pay for today’s investment in developing new software and systems. It is how these large companies justify their unique bets on the Smart Factory. This rationale may be valid for technology vendors, but the economics of a rip-and-replace approach is not feasible for factory owners. Big ticket analytics items may simply be unattainable in the short-term.
The new reality of the Smart Factory is almost identical to the old reality, and most of the building blocks for Asset Maintenance will remain unchanged for the foreseeable future. Putting aside vendors’ sales materials and product demos, these elements of Asset Maintenance will not be displaced anytime soon.
The Role of Rule-based Asset Monitoring in the Hybrid Smart Factory
Admittedly, the traditional legacy systems that use SCADA data to monitor asset performance have some limitations: the human inability to track all big data produced by numerous sensors in a factory and the reliance on humans to set manual control thresholds.
Furthermore, rule-based monitoring only tracks control breaches, and does not detect abnormal sensor data behavior within the prescribed control thresholds.
At the same time, there are critical physical signals such as a throttle position or a coolant temperature that still need to be monitored by factory maintenance technicians for an asset-specific reasons.
The benefit of using Big Data and Machine Learning for Asset Maintenance is that it can provide continuous real-time monitoring of all the sensors in a factory. In the Hybrid Smart Factory, machine learning will be used to detect abnormal and correlated patterns of sensors behavior to identify machine degradation or fault before they occur. Traditional asset monitoring systems will still be needed for high priority sensors that need monitoring for specific strategic/operational reasons.
The Role of Preventive Maintenance in the Hybrid Smart Factory
No matter how “smart” the factory, Preventive Maintenance (PM) cannot be completely replaced by artificial intelligence. Factory maintenance staff or third–party vendors with service agreements are responsible for conducting actual (scheduled or preventive) maintenance and to physically check on core systems and asset parts.
Preventive Maintenance is gradually becoming part of the Asset Maintenance mix, and that represents an opportunity to apply Machine learning to reduce both scheduled and unscheduled downtime.
Machine Learning algorithms analyze vast amounts of sensor data and detect abnormal behavior. The Artificial Intelligence algorithms identify the baseline for acceptable deviations from normal sensor behavior. When there are variances from the number of “acceptable” deviations, an alert of emerging degradation or fault is generated and sent to the facility maintenance staff.
With applied machine learning, scheduled maintenance tasks will be reduced, and resources can be used in other areas. Over time, when factory owners experience the impact of machine learning, they will lower requirements for double and triple redundancies. Reductions in Preventative Maintenance are likely to free up budgets for other types of investments.
The Role of Asset Simulation (a.k.a Digital Twin) in the Hybrid Smart Factory
There has been much excitement about the Digital Twin in the trade press. The idea that a factory can create a virtual clone of a machine asset that can be monitored in real time is appealing to industrial plants that have embraced the Smart Factory vision.
At the same time, there are significant hurdles to deploying the Digital Twin. The most significant challenge for incorporating the Digital Twin into their factory environment is the cost associated with implementation. Even without taking into consideration the software licensing fees to build the Digital Twin, accurate blueprints are needed to re-create a virtual model of the physical machine.
One needs to budget for an army of highly paid external consultants, big data scientists and design technicians who will work in tandem with the factory’s facilities engineers.
The Digital Twin can provide real time and valuable insights into the performance of a machine. Due to current limitations, the Digital Twin concept is not scalable. At best, it can be used for some high priority and high price machines. For many production verticals and facilities, the Digital Twin concept is not economically viable.
The Role of Automated Machine Learning for Asset Maintenance in the Smart Factory
Automated Machine Learning is a quantum leap in the world of data science because vast amounts of sensor data can be analyzed in real time. In Automated Machine Learning, AI is responsible for choosing and applying machine learning models to datasets. By using AI, and not data scientists, to control the model application process, Automated Machine Learning improves the speed and accuracy of data insights.
In SKF Enlight AI’s Automated Machine Learning solution, the approach is agnostic in terms of asset or sensor type, which means that abnormal data patterns can be detected for all factory machines.
What does this mean for the factory owner? In real time, emerging degradations of any machine part are identified in advance of a shutdown. Whereas the rule-based system will monitor just a few sensors, Machine Learning algorithms can analyze vast amounts of sensor data. The result is the ability to efficiently monitor all the assets in a production facility.