Select Page

Presenso CTO, Deddy Lavid addresses Joe Barkai’s concerns about Predictive Maintenance.

When Presenso was founded three years ago, Industry 4.0 was still in the planning phase and few industrial plants had direct experience with implementation.

Now, as we move from board-level strategy to shop floor deployment, new technologies are being tested in real-world scenarios. Industry observers are asking a simple question: Will the performance of these solutions match the hype?

This article will address some of the hard questions that respected industry analyst and author Joe Barkai asked in an article that appeared on his blog, titled Predictive Maintenance: Myths, Promises, and Reality.

#1. One Model to Rule Them All
Claim: The Digital Twin’s “focus on individual as-maintained unit configurations also highlights a potential concern: if over time, assets drift to the degree they are no longer similar, the ability to conduct any type of broad installed base analysis is impeded.”

Our Response: We agree with this assessment. Gartner has selected the Digital Twin as a top strategic technology trend for the last two years. However, the market response has been tepid, as evidenced by GE’s recent struggle with its Predix solution offering.

The Digital Twin is based on the development of a completely accurate virtual clone of the underlying machine asset. In practice, generating this clone is extremely challenging because the underlying blueprints of the physical machine are typically not accurately maintained. Furthermore, specifications for performance are imprecisely documented. Over time, these are ignored as assets are used beyond the original design limits. Finally, factory technicians are required to “teach” the Digital Twin Machine Learning algorithm. This is often impractical given resource constraints and other time commitments.

#2 It’s Magic: Machine Learning Algorithms Do All the Work
Claim: There is hype that “AI-based algorithms … do all the work on their own. They remove signal noise and data outliers, and smooth data just enough so no key features are lost; they identify the best-suited analytic algorithm, and provide highly accurate data trending and failure prediction.”

In reality, early adopters discover that scaling an AI-based predictive maintenance solution is difficult, “as the models must be validated for a much broader range of product configurations and applications, and be able to adapt to changes induced by cyclical changes and wear and tear.”

Our Response: The concerns about AI-based algorithms are valid but can be mitigated if Automated Machine Learning (AutoML) is incorporated into the Predictive Maintenance solution.

AutoML is an important development in the data science discipline. It applies the science of Machine Learning to the practice of Machine Learning. The objective of AutoML is to replace many of the repetitive and time-consuming tasks of the data scientist, including data preprocessing, algorithm selection and hyperparameter optimization.

The field of Predictive Maintenance has two major and direct benefits.

First, replacing human tasks with AI-driven decision making speeds up the algorithm selection process. From a scale and implementation perspective, this enables cheaper and faster adoption within an industrial facility. The dearth of qualified data scientists in the market has served as a catalyst for the adoption of AutoML.

Second, with AutoML the model is built rapidly. As a result, data leakage can be detected early in the modelling lifecycle. The practical implication is that as the Machine Learning pipeline development is optimized, the model becomes more accurate.

Although still a relatively new field, the use of AutoML can accelerate Machine Learning processes and reduce the need for human intervention. These advances in data science directly address legitimate concerns about the scalability and adaptability of models highlighted by Joe Barkai.

#3 You Don’t Have to Be an Expert to Make Sense of the Data
Claim: Predictive Maintenance cannot be implemented without an understanding of the underlying machinery. This is because “operating signatures of rotating and reciprocating machinery varied by configuration and duty cycle, and changes over time due to wear and tear, making sense of data patterns and prescribing the appropriate response require more than mere statistical data analytics.”

Our Response: There are two fundamentally different approaches to Machine Learning – Supervised and Unsupervised.

With Supervised Machine Learning, the algorithm “learns” based on human instruction. The learning algorithm requires labels of anomalous and normal machine asset conditions. Then, when the algorithm encounters new data, this data can be classified according to the pattern on which it was trained.

With Unsupervised Machine Learning, the algorithm does not have to be trained using the physical blueprints or knowledge about the process itself. Using Advanced AI, the algorithm recognizes data patterns without receiving prior training on the underlying asset. Vast amounts of data are analyzed and the algorithm itself generates the label.

Unsupervised machine learning for industrial analytics

 

 

 

 

 

 

 

Source: Presenso

Why is this significant? If the algorithm itself can detect changes in configuration, wear and tear, etc., there is no need for humans to replace or recalibrate the learning model because this is done automatically.

#4 Good Vibrations
Claim: “A signal signature will vary from one unit to another and will keep changing over time. And it’s easy to overlook the details that separate the real world from the lab. For instance, identical units installed using different mounting techniques on different foundations are likely to exhibit different waveforms.”

Our Response: From Presenso’s inception, the concern raised has been reflected in our approach to Machine Learning. As we wrote above, the core of Presenso’s Machine Learning methodology is that the algorithm itself continuously adapts to changes in the asset and does not depend on human input. The combination of AutoML and Unsupervised Machine Learning is how Presenso has addressed this issue.

#5: If You Build It, They Will Come
Claim: Predictive Maintenance solutions have not been well-received by service technicians because they are typically “tedious, inflexible, and overly authoritative, and not sufficiently tuned to their needs, habits, and workflows.”

Our Response: We have not surveyed enough competitive offers to reach a definitive conclusion about this topic.

At Presenso, we have prioritized the development of the UI. UI is more than an “attractive” or even “intuitive” interface. We recognize that Machine Learning algorithms generate complex output and that unless they are simplified for the end user, the full potential of the system will not be utilized.

Ultimately, the simple visualization of insights allows the technician to absorb complex information in a way such that decisions can be made in real time without first-hand knowledge of the underlying Machine Learning science.

Summary and Conclusion
McKinsey & Company predicts that Industry 4.0 and Predictive Maintenance can unlock hundreds of billions of dollars in value for the industrial sector. The magnitude of this opportunity creates a financial incentive for the investment community to fund emerging players in this space.

Each company that wishes to compete in the emerging Predictive Maintenance category will need to address the legitimate concerns that this article outlines.

If you wish to learn more about Presenso’s approach toward IIoT for Predictive Maintenance, click here to attend a webcast about our solution.

Like this article? Please share it using these icons....Share on FacebookShare on Google+Tweet about this on TwitterShare on LinkedIn
Deddy Lavid

Deddy Lavid

Experienced R&D Manager. Recognized expert in the field of Machine Learning and Big Data architecture. His work spans the full spectrum from researching isolated data problems to building complex production systems. At Rafael, he led a team of algorithm developers in large Software projects of national importance. Holds honors M.Sc. computer and information science.