Engage key stakeholders

Predictive Maintenance Needs Employee Buy-in

Stakeholders will have varying concerns about introducing AI-driven Industrial Analytics within the factory. Familiarizing yourself with these objections can allow you to understand the reasoning and motives driving these reservations, and to craft an onboarding strategy that addresses each stance.

The Importance of Addressing Internal Objections

Industry 4.0 and Maintenance 4.0 are considered the fourth industrial revolution. The digitalization of the industrial world brings massive and disruptive changes that pose a significant risk. Plants must balance the risk involved in rushing too quickly to invest in unproven technologies. However, waiting on the sidelines and failing to act also entails risk.

A clear understanding of internal objections gives those tasked with affecting change a view into the challenging terrain ahead. We cannot institutionalize change until key stakeholders with executional responsibilities are acknowledged.

The purpose of this document is to list some of the common objections to Industrial Analytics deployment and to provide guidance on how to frame the conversation to address these concerns.

Overcoming Common Objections to Implementing Industrial Analytics

One of the reasons that many Industrial IoT projects get stuck in “pilot purgatory” is that many of the Maintenance 4.0 solutions are still in their nascency. There is no one common roadmap or standard for deploying Industrial Analytics.

Perhaps the price that is paid for the hype surrounding Industry 4.0 is the confusion on the part of those responsible for its implementation.

Artificial Intelligence Is No Match for Human Experience

The following guidance can be used to discuss these common objections:

An area of resistance likely to emanate from the plant floor is that experienced Reliability and Maintenance professionals are intimately familiar with plant equipment and that a machine-driven algorithm cannot match their insights.

The underlying assumption supporting this position is that intuition, common sense and context trump insights gained by analyzing data patterns.

Here are ways to overcome this objection:

  • Reduced Cost of Big Data Analytics Data Storage and Computational Cost Reductions Enable Widespread Application of Machine Learning
    One of the biggest drivers of Machine Learning-based Predictive Maintenance adoption has been the significant reduction in the cost of capturing, storing, transporting and analyzing sensor-generated data.
    Long-term favorable cost trends have enabled Machine Learning to be broadly applied across an entire plant’s asset base.
  • Scalability and Asset Coverage While human expertise is limited to a relatively small number of machines with historical significance, Machine Learning can be applied exponentially. When vast quantities of data are generated from sensors throughout a plant, Industrial Analytics is used to identify correlative patterns of anomalous behavior and can alert plant technicians to the root cause of evolving failure that is undetectable by human logic.

Practitioner’s Tip

When addressing this concern, we should first acknowledge the contribution from Reliability and Maintenance professionals, even if they are constrained by human limitations. For many, the move to industrial analytics is considered disruptive, and plant employees’ concerns should be recognized. The best approach for evangelizing Industrial Analytics is to position it as a complementary tool that enhances the performance abilities of front-line technicians.

Current Preventive Maintenance Programs are Effective

An industrial plant that has built robust Preventive Maintenance practices may consider advanced Industrial Analytics to be unnecessary. The logic is simple: We use OEM best practices to service and inspect critical industrial equipment which prevents asset degradation.

Here are ways to overcome this objection:

  • Strive for an optimal mix of maintenance strategies Although Preventive Maintenance is a critical component of any maintenance program, time- or usage-triggered maintenance activities have limited benefits. About 80% of plant machinery degradations or failures are not based on age factors and occur for unknown or random reasons. Costs are associated with over-maintenance and under-maintenance – and both should be avoided. Rather than viewing Industrial Analytics and Preventive Maintenance as a trade-off / zero-sum game, industrial plants must find the right balance between the two.
  • Preventive Maintenance has unintended consequences A downside of Preventive Maintenance is that, in many scenarios, it leads to asset degradation and failure. Simple human error, such as misreading dials or using the wrong type of repair equipment, can result in damage.In one study of fossil-fuelled power plants, the majority of unplanned maintenance outages occurred less than a week after a planned outage (i.e., 1,772 of 3,146 maintenance outages occurred after an outage).

Practitioner’s Tip

Position Industrial Analytics to plants with strong maintenance practices as incremental. Industrial Analytics should fit into existing maintenance practices and take into consideration existing expertise and best practices.

We Lack Expertise in Machine Learning

In research conducted we conducted with Emory University, the number-one factor inhibiting the deployment of IIoT Predictive Maintenance was a shortage of data scientists.

This concern forms the basis of a common objection to implementing solutions that require this expertise – that industrial plants’ inability to recruit professionals will limit their ability to deploy Machine Learning-based applications.

Here are ways to overcome objections based on a lack of Machine Learning expertise:

  • Industrial analytics tools are not dependent on plant-level expertise If plant-level employees are considered a bottleneck for Industrial Analytics deployment, the underlying assumptions are likely flawed. Some of the first-generation Industrial Analytics solutions required the input of plant-level expertise to train learning algorithms. However, this is not a scalable model – reliability technicians cannot become “citizen” data scientist to support Machine Learning-based Predictive Maintenance.
  • Automated Machine Learning compensates for labor supply constraints Automated Machine Learning or AutoML is one of the most significant advances in the data science discipline in the last couple of years. Many of the manual data science processes, suchasdatapre-processing and algorithm model selection, are labor-intensive and can therefore impede the widescale adoption of Machine Learning-based Predictive Maintenance. With AutoML, manual data science tasks are executed by Machine Learning algorithms. The result is that few data scientists are needed for the deployment of Industrial Analytics. As AutoML processes become more widely adopted, the labor supply constraints become less of a significant blocker for large-scale deployment of Industrial Analytics solutions.

Industrial Analytics Not Suitable Based on Asset Age

One objection to Industrial Analytics is that assets experience predictable failure patterns due to age-related factors. Therefore, there are times within the asset life when Industrial Analytics cannot be justified.

As depicted below, there are both age- and non-age-related failure patterns.

For instance, the so-called Bathtub Curve suggests that assets are most likely to fail at the beginning and end of an asset life. Due to defections in new machinery, the initial ownership period is considered the infant mortality stage or burn-in where failure rates are relatively high. This is followed by a steady state of normal asset life. At the end of life, there is an increase in the failure rate due to asset wear-out.

Objections based on asset life can be addressed in the following ways:

  • There is no need for Predictive Maintenance for certain asset age categories Research conducted by United Airlines suggests that only 11% of asset failures are age-related and that 89% are random. Although certain patterns of asset failure can be depicted based on historical data, the learnings cannot be applied to an individual asset. The vast majority of asset failures are random and not time- related. Predictive Maintenance can be used for assets across their lifespans, including infancy, steady state and end of life.

Other Common Objections

Below is a list of common objections that our sales team has provided, along with basic guidance in addressing the issues raised.

  • Industrial Analytics is expensive Every new solution has a cost. The most important consideration is the value relative to the investment. The core question is: What are the financial benefits of reducing the incidence of Unscheduled Downtime? We recommend conducting a Business Value Assessment based on expected incremental revenue from increased production throughput and cost savings for lower O&M expenditures.
  • Machine Learning does not work It is true that not all Machine Learning solutions are equal. Our recommendation is to conduct a pilot / Proof of Concept and to evaluate each solution’s predictions of failure versus the actual log data. Purchase decisions should be data driven.
  • The Digital Twin is too difficult to deploy on brownfield assets The Digital Twin has both advantages and disadvantages. Although it is expensive and complex to deploy on brownfield assets, there could be an overriding business reason to justify the development of a virtual clone of a physical machine. Purchase decisions should be based on the business value relative to the cost and time to deploy.
  • Industrial Analytics is not a priority If senior management does not consider Maintenance 4.0 to be an organizational priority, Industrial Analytics will not be a priority either. A different (and more likely) scenario is that a high-level strategy for Maintenance 4.0 has been devised but not fully articulated throughout the organization. The result is that individual stakeholders are unaware of the broader importance of Industrial Analytics and do not recognize its significance. The best way to address this issue is to get clarification on the company’s overall Maintenance 4.0 goals.
  • There has been a negative past experience with Industrial Analytics There are numerous reasons why an Industrial Analytics deployment may have failed. Instead of using past experience to justify inaction, consider the specific learnings that can be gained from postmortem analysis. Start with an understanding of whether the Industrial Analytics solution failed due to a software flaw or to a lack of employee training. Using this knowledge, potential roadblocks can be identified upfront.
  • There is a lack of infrastructure to deploy Industrial Analytics Implementation of an Industrial Analytics solution is dependent on the following:
    • Embedded sensors that generate data about machine performance
    • Connectivity that enables the capture, transport and analysis of the sensor data

    Organizations that do not capture sensor-generated data should consider a medium- to long-term plan that addresses infrastructure issues. If Big Data is an unused and wasted asset with a significant economic upside, at the very least, organizations must align the extraction of value from this data with the overall strategy and invest accordingly.

Summary and Conclusions

A range of objections to Industrial Analytics will be articulated by various stakeholders within an organization. Some of these objections are based on valid scenarios, whereas others are based on incorrect reasoning and assumptions.

In either case, before one addresses the objection, it is important to understand both the rationale of the argument and the motivating factors.

Decisions about new investments should be based on an accurate assessment of the current state as well as realistic projections of long-term value. A deep understanding of stakeholder objections can provide insights into both.

SKF Enlight AI

Industrial plants generate terabytes of process data. SKF Enlight AI is a SaaS Predictive Maintenance solution that uses Automated Machine Learning to identify emerging asset failure patterns within this data. It provides early warnings and sensor-level intelligence to help avert unplanned downtime and meet production goals. For more information on how SKF Enlight AI can improve performance and reliability, click here.

Leave a Reply

Your email address will not be published. Required fields are marked *