By: Beth Crane, VP of Data at Sight Machine
Among manufacturers, unplanned downtime related to system maintenance is a common source of decreased productivity and profits. As access to data increases, manufacturers are aiming to leverage sophisticated data models to predict maintenance requirements and optimize the timing of downtime. When successful, these predictive maintenance efforts can significantly reduce the costly unplanned downtime associated with machine failures.
As the VP of Data for Sight Machine, I’ve worked with numerous global leaders on their transition to becoming data driven manufacturers. What I’ve discovered is that many manufacturers are surprised by the difficulty and cost associated with implementing predictive maintenance projects.
That’s not to say these initiatives can’t be successfully executed, however, to truly be successful, manufacturers need to think about predictive maintenance within the context of a broader set of data-driven process reengineering efforts that can benefit their business.
Why this is hard
Manufacturers are often surprised at the cost and difficulty of the predictive maintenance projects. These projects require predictive models built on your production process that enable predictions of possible machine failure. If the underlying processes have too much variability, the predictive models will be unable to accurately and reliably forecast the failure of parts or equipment.
In addition to developing and implementing the models, these predictive models will require constant monitoring and tuning to ensure they continue to accurately represent the physical environment. This ongoing maintenance can also contribute to the significant costs of predictive maintenance efforts.
Taking a comprehensive approach
At Sight Machine, we’ve learned that the most successful manufacturers are those that build capabilities for data-driven process reengineering before embarking on predictive maintenance efforts. There are three primary capabilities that are required:
1) System stability: Excess variability is often associated with one or more material and equipment parameters. In many situations, assessing and attaining stability can be more valuable and more cost effective than developing a complex predictive model for preventive maintenance. It is not uncommon that addressing an issue related to variability in the system eliminates the need to have the predictive model at all.
2) System monitoring: Monitoring is a foundational capability that enables you to detect variability in your system. Once detected, it is necessary to evaluate whether the variability is random or impacting the production process. To the extent possible, variability can and should be removed to reliably produce products to specification. Without monitoring capabilities, manufacturers will be unable to determine when their processes begin to lose stability and therefore, when their predictive models become ineffective.
3) Real-time process reengineering: Most manufacturers lack the processes to quickly act upon any data-driven insights. Accurate monitoring is only as good as the proactive response by the manufacturer to any potential issue detected. Whether alerted to variability or to a maintenance need, responding to predictive maintenance forecasts requires you to define the workflow to address monitoring alerts or predictions: Who needs to know about this? How are they notified? What information is required to effectively and efficiently respond to a predicted failure or maintenance need? And what do you need them to do once they are notified?
I’ve discovered, that in the process of building these capabilities, many manufacturers determine that they can achieve significant process improvements without the need for an expensive predictive maintenance models. Even in situations where predictive maintenance models do make sense, investing in the tools to build system stability, monitoring, and real-time process reengineering will ensure that predictive maintenance efforts have the required prerequisites to deliver long-term impact.
For example, catastrophic machine failures might be associated with poor adherence to SOP or maintenance protocols. These types of issues can often be addressed by training and organizational policies or protocols that encourage adherence to the defined protocols. This is usually much cheaper than building and maintaining models required for successful predictive maintenance. Setting up visibility into data that allows operators to monitor variables and change processes as needed could provide much of the benefits without the cost of a predictive model.
So when does a predictive model for preventative maintenance make sense? In situations where the prerequisite work to create a stable system has been completed, reliable and accurate monitoring is in place, and organizational policies and processes are established for effectively responding to alerts that maintenance may be required.
With advancements in the ability to easily access machine data and apply sophisticated analytics, predictive maintenance projects can be an effective tool for reducing the overall cost of unplanned downtime. By looking at predictive maintenance efforts within the context of a broader data-driven process reengineering foundation, manufacturers can build a foundation to ensure these projects deliver on their promise.