Seite wählen

Leider ist der Eintrag nur auf Amerikanisches Englisch verfügbar. Der Inhalt wird unten in einer verfügbaren Sprache angezeigt. Klicken Sie auf den Link, um die aktuelle Sprache zu ändern.

Although IIoT Predictive Maintenance is still in its nascency, it is a theoretical concept. Today, sensor data generated by industrial equipment is providing real-time alerts about evolving machine failure. Many industrial plants are piloting IIoT Predictive Maintenance and we are witnessing partial rollout by early adopters. For this article, we set aside the different applications of Artificial Intelligence, such as Presenso’s Automated Machine Learning methodology and the Digital Twin.

As we move toward systems of algorithm generated failure alerts, it becomes the responsibility of Asset Maintenance professionals to remediate evolving degradation before machinery shutdown occurs. In this article, we have solicited expert feedback about the following question: How must current O&M practices adapt to Machine Learning Predictive Maintenance generated alerts?


Dan Arsenault CMRP

Dan Arsenault CMRP, Reliable Asset Management Services

The move toward IoT Machine Learning Predictive Maintenance is inevitable and will likely happen much faster than most maintenance people believe. Hence, there will be knowledge and understanding gaps that should not be inevitable. I believe it will be many years before machines are self-repairing, so qualified technicians will be in demand for a long time to come. However, traditional skills and behaviors will not be able to cope with the demands of smart systems.

The biggest shift will have to occur in the mindset of the operations and maintenance staff. They will have to be trained on their processes and systems in detail. They will have to understand what is being measured, why it is being monitored and what to do in the event of an alarm or alert. It is far too common for both operators and maintenance staff to hit the “reset” button and ignore the underlying issue. Systems that record this data should be monitored for alarm “reset” activity and then immediately dealt with in a disciplined manner.

Companies should be looking to modernize their staff with young graduates who have Electro-Mechanical diplomas in Robotics, Automation, Mechatronics or similar disciplines. The technicians should be able to understand and solve issues with the physical aspects of the machines, such as hydraulics, pneumatics, and power transmission systems, as well as the machine control systems. Remote I/O devices talk to PLC’s and microprocessors, loaded with programs and menus, which in turn communicate with databases and servers. Comfort with computers systems is now a prerequisite for industry technicians.

Equally important are the basic concepts of Reliability, OEE and Lean Sigma methodologies. Smart sensors are like augmented reality; they see, hear and feel abnormal machine conditions long before our senses notice that something is happening. So, understanding the P-F curve and how it describes failure, as well as understanding the different failure modes for each machine component, will help technicians plan their adjustments or repairs long before failure occurs. Risk Assessments and FMEA’s should be standard maintenance practices.

Maintenance leaders must urgently recruit, develop and grow their staff to meet future needs today!


Jim Humphries

Jim Humphries Principal at Performance Dimensions

Certainly, new skills, knowledge, and approaches will be required to support and capitalize on the new technologies. Other than machine learning and AI, the tools have been around in industrial facilities in some fashion for decades. However, their limited application has been the result of narrow functionality, relatively high costs, and bounded successes. This has limited not only application but also associated learning curves. The key to broad application and acceptance is low cost and ease of application.

Although the cost of sensors, software and hardware is an important consideration, the limiting cost drivers have been primarily component installation and integration. Hardwiring the many sensors required for broad functionality is costly in new facilities and even more difficult in running facilities. The software to expertly address multiple failure modes has been almost non-existent. This has limited the scope of applications to narrow sets of equipment and failure types except for those users willing to take on research and development. To take on the cost issues going forward, progress meeting wireless and ethernet over power challenges in industrial environments must continue.

New software engines to capture, integrate and analyze a broad array of sensor, operations and maintenance data are essential. Because of the variety of equipment and failure modes, the trend toward niche suppliers providing specialty systems for categories of equipment such as rotating equipment, instrumentation and electrical distribution will continue. Software to integrate these systems with other operations and maintenance data and to apply broad learning techniques is essential.

Formal and informal standards for sharing condition data, including status determinations, are necessary to drive the integration as well as the sharing of niche processing demands across the internet. Although standards will certainly promote ease of application, much more is needed. A plug and play approach toward component and failure mode additions in a facility installation will create the benefits of modular architecture. Open and clearly documented libraries for automated diagnostics of common component failure modes will aid in the understanding and, ultimately, the implementation of required maintenance actions. The ability to use the libraries as models in tools to build new diagnostic routines will be quite helpful. When AI routines drive diagnostics based on correlations or predict failure timing, clarity in the reporting of rationales will also drive the acceptance of suggested actions. When suggested actions are inappropriate, the ability to tune simply and easily with transparency will support continuous improvement and enhanced acceptance.

Clearly, answers to the needs identified above will evolve over time. Users will want to easily enhance incrementally rather than make wholesale replacements of software and hardware. With this, the possibilities and successes will continue to grow as these new tools become accepted elements of excellent operations and maintenance.


Deddy Lavid

Deddy Lavid, CTO of Presenso

It is inevitable that Failure Detection alerts are becoming system generated and that some of the supporting maintenance processes will also be automated. As a first step, I see the automation of “no-regret” maintenance activities. For example, certain processes can be activated with minimal risk of over-maintenance or incorrect maintenance. The lubrication of a machinery part is a low-risk activity that can be triggered by a Failure Detection alert and be automatically executed. The second step is that work orders and schedules will be scheduled automatically based on evolving failure.

I do not wish to imply that repair work will be automated and that humans will not be part of the process. In the first phases of IIoT Predictive Maintenance, O&M professionals will receive algorithm-generated guidance that prioritizes maintenance activities. It is reasonable to infer that, over time, automation and robotics will play a larger role in the execution of the algorithm-generated Failure Alerts. The roles within O&M are likely to evolve as more Industry 4.0 technologies are integrated into Predictive Analytics.


Mike Spence

Mike Spence, Senior Reliability Engineer at Refining NZ

The ability to use machine data to generate work orders automatically has existed for decades. Yet, many successful companies survive without leveraging this capability.

If a simple approach is still unjustified, what is the value of IIoT Analytics? The bottleneck is actually the development, deployment, and implementation of corrective/preventive/proactive strategies. Most CMMS/work execution systems are inflexible regarding the routing and scheduling of non-invasive, flexible, low-cost (but high-return!) activities.

In my opinion, this is the elephant in the room! The best approach I have yet seen and worked with for these activities is a package by Generation Systems called LubeIt (no affiliation), yet it is still a relative antique – and unique. No matter how it is managed, radical changes in human and automated maintenance deployment are required to leverage the advantage of IIoT analytics.

As a Reliability Improvement engineer, I foresee that the IIoT Predictive Maintenance model will ultimately be leapfrogged by a rejuvenated Continuous Improvement model. Managing known problems is usually less effective than solving them. The intriguing aspect will be whether reduced-cost maintenance activities (via maintenance robot/drone deployment) become more worthwhile than solving/preventing problems. Self-regenerating pipelines, anyone?


Giuseppe Padula

Giuseppe Padula, Professor of Advanced Design, University of Bologna

The large utilization of IIoT devices within the manufacturing environment, combined with the rapid adoption of AI and digital twin systems fed by sensor data, creates as a consequence the migration of traditional alarm systems, controlled by deterministic events, toward a mixed system, controlled by a combination of deterministic and predictive ones. In the latter case, the efficient grouping and sequencing of the two types of alarms would help O&M personnel organize assistance activities in a more rational and effective way, thereby avoiding unexpected events and leveling the production flow.

Who will be responsible for deciding how to transform the predictive events into actionable activities?

Companies may choose to either leave those decisions to an intelligent system or assign them to the person in charge of O&M, who may then decide whether to execute or postpone the actions proposed by the system itself.

Depending on the extent to which the decisions will be automated, the role assigned to the O&M technicians will be clearly different, as will the relevant tasks to be performed in order to deal with those new intelligent systems in different degrees of involvement.

However, operators will play an important role even before the full implementation of the Predictive Maintenance systems –that is, during the phase of machine learning.

An effective and valuable training of AI systems requires that skilled O&M professionals, who have the necessary experience to build predictive manufacturing models, be deeply involved from the beginning.

The Machine Learning phase can be used to create the new interactive framework between O&M professionals and AI systems, which later would evolve into the operative framework where technicians and the AI system will collaborate in supervising a more predictive and less deterministic manufacturing environment.


Robert van der Veur

Robert van der Veur, Project Manager -IoT, Business Intelligence & Analytics, Sogeti Nederland B.V.

Machine failure can shut down a production line for weeks. Depending on the severity of the failure, the total cost impact can easily reach millions of dollars per week.

O&M’s across different industries are, by design or default, at different stages of maturity. Some may be running scheduled maintenance checks based on estimates or OEM recommendations, while others may utilize statistics-based programs individually tailored to each fixed asset.

Still, others are already employing continuous monitoring technologies of their assets but may be monitoring only the outputs of the data, rather than leveraging advanced Machine Learning Predictive Maintenance models.

Creating an early warning system using the Machine Learning Predictive Maintenance model to prevent machine failures could detect the first signs of failure, providing time to plan and schedule the replacement or restoration of a failing part without interrupting production with most failures.

Adopting these advanced maintenance models has an impact on these O&M’s. For example:

  • O&M teams that manage equipment reliability with time-based preventive maintenance programs are still not aware that only 11% of failures are age-related and that the rest of the failures are random!
  • The number of inspections can be reduced. Sensors are monitoring the fixed assets 24×7 and can give real-time status updates.
  • O&M teams will be more proactive, preventing many equipment failures. Their corrective maintenance work will be greatly reduced. With that, instead of replacing the entire piece of equipment due to critical failure, a repair can be made prior to failure. The cost will be minimized to the price of the component and the labor needed for the repair.
  • O&M teams will see increased efficiency in terms of employee time. Identifying the precise repair task needed to correct deficiencies, as well as the parts, tools, and support needed to correct the problem, can dramatically increase effective “wrench time.”
  • With less machine failure, operators will save more.
  • O&M teams will plan their work based on the alerts generated by a Machine Learning Predictive Maintenance system.