Since 2016 Petasense has provided IIoT-based condition monitoring solutions to various industrial segments. Over that time we have been able to gather substantial, first-hand experience from pilot programs and full-scale implementations. As both interest and adoption of this technology grows rapidly each year, we would like to share some of our learnings from actual projects to help you navigate the most common challenges in your own program.

The practical benefits of incorporating IIoT solutions within industrial reliability programs is manifold. Many publications have discussed in depth the reasons companies choose to complement existing legacy condition monitoring programs (1, 2). Some of the top reasons include leveraging human resources, allowing greater coverage of assets (both spared and balance of plant), increasing personnel safety and providing near real-time monitoring of assets.

As with all new technologies, there is a growth curve consisting of both technology maturation as well as becoming knowledgeable about the correct use and implementation of said technologies for achieving the desired results. This article will shed some light on the most common pitfalls we see with IIoT-based Condition Monitoring implementations and provide some areas of considerations to help you achieve optimum results.

 
1. Unclear or Undefined Objectives

While having clear objectives is important in any project under consideration that is competing with other ideas for budget funding, it is even more so with new technologies. This is true not only for success and failure criteria, but also because in many instances companies will have some type of condition based monitoring program in place already, which the new technology will be compared to.

The top objectives we’ve seen for implementing IIoT-based condition monitoring include:

i. Supporting digital initiatives for modeling and automation of plant production. In many cases plants lack sufficient data (coverage and types of data) to adequately perform advanced analytics and modeling

ii. Supporting insufficient human resources. Resulting from the aging workforce retirements or lack of trained personnel to collect, analyze and advise

iii. Addressing the challenges of achieving adequate asset monitoring. Especially for assets that are remote, spared, run in batch processing, or located in areas off limits to personnel

iv. Addressing the requirement to obtain more frequent monitoring. Often to ensure detection of abnormal conditions to prevent safety hazards or regulatory issues.

v. Avoiding the excessive costs of hardwiring or scaling a program with legacy systems

Identification of why you are considering implementing wireless systems and selecting the right assets to monitor are essential in meeting successful pilot criteria.

 
2. Lack of Coordination Across Departments

Well over 50% of the pilot deployments we see are initially funded by IT departments, typically through digital transformation initiatives. While this allows for the trial of new technologies outside of traditional operating budgets, it can be detrimental if not intimately tied to the ongoing concerns of maintenance, reliability and production.

Typically a digital initiative is concerned with the 3 V’s of data; Volume, Variety and Velocity; as well as security, bandwidth and how easy it is to ingest or export data through various integrations and APIs (Application Programing Interfaces). The above list is not fully comprehensive, but represents the majority of the questions we receive with the intention to support the data lakes and advance modeling and analytics that businesses are interested in.

However, maintenance and reliability teams also use the technology daily to achieve their objectives of reducing costs, increasing uptime and improving safety. This requires that the system be able to do more than aggregate data and display colorful charts and dashboards. The system needs to also have the capability to perform advanced diagnostics and to allow resident experts to customize the system based on local knowledge as needed.

In our experience programs will not scale past a pilot stage if there is not sufficient depth and breadth with the system’s functionality (more on this later). It is important to recognize that the legacy systems in use today may be well over 20-30 years in development and have quite a bit of capability. Maintenance professionals are not willing to sacrifice a great deal in this area just to try something new. Finally, one should not view this as a binary decision in all cases. In the majority of cases, wireless technologies are best applied as a complement, not a total replacement, for legacy systems.

 
3. Incorrect or Inadequate Sensors

Another important aspect of implementing a successful IIoT-based condition monitoring system is sensor selection. There has been amazing progress in MEMs vibration sensor technology with respect to frequency response, linearity, power consumption as well as wireless communication methods. However, our testing has shown that there are too many examples still of sensors that do not perform to their published specifications. This becomes more crucial when selecting vibration sensors, which vary tremendously from each other in terms of capabilities.

The key sensor specifications to look for (and possibly test to confirm) include:

i. Frequency Response: What are the minimum and maximum frequencies the sensor is capable of measuring with + or – 3 dB linearity. Will the sensor selected be able to adequately detect fault types with sufficient warning.

ii. Triaxial Vibration: Most, although not all, wireless sensors today are capable of triaxial measurements (versus a single axis). This is essential if one is working remotely for diagnostics. Some sensors do not provide the same frequency response on all three axes, while others, when tested on a lab certified shaker table, do not perform to their stated specifications. Care should be taken to ensure that your sensor performs as required for the application.

iii. Vibration Spectrum and Waveform (Not just overall RMS): Some sensors only provide overall measurements (the equivalent of a check engine light) while more advanced sensors include dynamic data (FFT and Time Domain) that allows for actual fault diagnostics and early detection. The lack of comprehensive data is usually due to a need for prolonging battery life or work within existing sensor-gateway communication limitations. A proper implementation must include both overall and dynamic vibration data, collected at least on a daily basis.

iv. Sensor Size and Footprint: If the sensor design is too large it greatly limits where the sensor can be installed. (More on this later when we discuss the importance of sensor location & installation)

v. Battery Type and Power Options: Some wireless sensors utilize non-replaceable or proprietary batteries. This is not only an environmentally poor design, but it also requires that the sensor be thrown away. It is a huge benefit for replacement batteries to be readily available, economical and replaceable by end users in the field.

 
4. Improperly Installed Sensors

This is one of the most common challenges we see and is one area where a simple error can undermine the entire program. We have observed that improper sensor installations often fall into the following categories:

i. The wrong type of sensor being applied: When this occurs the sensor cannot adequately detect with enough lead time the types of faults you are looking for. For example, a vibration sensor with a low frequency response will never be able to detect early-stage bearing issues.

ii. Improper sensor placement: Placing the sensors on sheet metal shrouds, too far from bearing structure or poorly mounting on motor fins without special fin mount adaptors leads to erroneous data.

iii. Inadequate quantity of sensors: Placing sensors too far from the vibration transmission path (typically near the bearing) will likely result in poor detection of faults.

iv. Improper mounting methods: mounting bases with stud fasteners or industrial epoxy is preferred.

Most of the above can be avoided when working with the proper application knowledge and subject matter experts. Unfortunately, setting up and operating a good condition monitoring program is not yet an internet or catalog business and should not be treated as such.

 
5. Lack of Software Functionality

Most providers of IIoT in the market today are less than 5 years old. Many of the initiatives were founded with innovators from the digital healthcare and telecommunications sectors. While this has injected some much needed technologies new to condition monitoring, it has also resulted in a lack of deep domain knowledge in many cases as it relates to asset health monitoring science.

Most applications today provide some type of HMI (Human Machine Interface), or Dashboard, to clearly and quickly alert the user about important events and program details. Given the tremendous amount of data that must be turned into actionable information, this is certainly necessary, but not sufficient.

For programs to be successful the application must also be able to support various analytic tools that should include both ML (Machine Learning) and AI (Artificial Intelligence)(3). In our experience the best results are obtained when they provide both “out of the box” support, to quickly get started, as well as allow for customization through direct expert intervention and supervised learning tools. Only in this way can false positive and false negative notifications be reduced to acceptable levels and accurate insights be generated.

One interesting area of innovation has been the introduction and rapid adoption of SaaS (Software as a Service) that leverages Cloud computing. This has provided many benefits such as reducing IT support costs, greatly increasing the availability of new features and remote access 24/7 at a global scale. However, the results above are best achieved performance wise when the application has been written from scratch with Cloud Native Architecture. Simply modifying legacy Client Server applications is fraught with issues when deployed at scale.

Finally, to be accepted by maintenance and reliability teams, the software application must provide the user with the diagnostic tools they have been accustomed to and utilize to perform in-depth diagnostics and root cause analysis. While the analytics have advanced tremendously in the last decade, it is still prudent to provide these tools in order to validate automated insights and enable a deeper dive to uncover root causes. Such tools enable proactive maintenance and support efforts to increase MTBF (mean time between failure).

 
6. Over-Reliance on Dashboards, AI, ML and RUL (Remaining Useful Life)

Condition monitoring has been one of the bright spots for Industry 4.0 and IIoT. Numerous industry surveys have all consistently ranked Condition Monitoring and Predictive Maintenance as one of the top 10, if not top 5 use cases (4).

Having said that, it is still considered to be in the early adoption phase on the curve. This means there has been quite a bit of hype around what wireless systems and automated analytics can and cannot do. This has created some healthy skepticism from maintenance and reliability experts when it comes to trialing these cutting-edge systems.

It is important to closely examine the sensor selection (covered earlier) along with the application at hand. Just as important is to understand both the benefits and limitations of the dashboards and analytics these programs provide. AI and Machine Learning algorithms enable the user to screen and filter the tremendous amounts of raw data available which need to be turned into actionable insights (in fact, it is the only practical way to scale). However, AI models are limited in the amount of contextual data and background knowledge available. Therefore we still recommend having a layer of human review before conducting corrective actions, at least until automated models can provide such recommendations with a high degree of accuracy and contextual awareness.

 
7. Considering IoT as a Replacement rather than a Complement to Existing Systems

One mistake we have seen repeatedly is the approach whereby the use of IIoT systems is being considered as a total replacement for other existing condition monitoring systems. This frequently results in pushback from existing users, but more importantly, can result in less than optimum reliability program results.

As mentioned previously, IIoT condition monitoring systems have many benefits over existing hardwired and portable data collector systems. They do, however, have some drawbacks. It is important to take these into consideration to make sure the program as implemented does not disappoint.

For example, these systems are not suitable (at least today) to serve as machinery protection systems. Most are not capable of monitoring and reacting within milliseconds. They are typically not designed for redundancy. They may have data latency issues. Advanced troubleshooting may require many more measurements than a wireless installed system can provide. In these cases, portable systems provide the necessary capabilities.

 
8. Inadequate Wireless Infrastructure

One particular challenge faced by all online systems, including wireless systems, is device reachability. This includes maintaining proper power (or battery levels), robust networks and any ensure communication of routers and gateways. We have also observed instances of plant maintenance inadvertently removing sensors during maintenance and not re-installing them or re-installing them incorrectly. No system will work reliably without the capability to observe and maintain system integrity. The good news is that many modern IIoT systems continuously monitor the system itself and will immediately notify users of performance issues such as poor connectivity.

 
9. Unclear team roles and Lack of proper training

This challenge is not confined solely to IIoT implementations, but is one of the very common obstacles we encounter with all projects and initiatives. It can be amplified by the nature of the wireless remote program in that there may be several entities that are responsible for the system. For example, technicians install and maintain sensors, vendors or SMEs may monitor the system notifications and reliability teams may be responsible for final diagnostics and disposition of alarms.

In many programs the personnel involved tend to change over time, particularly when shifting from pilot to production scale. It is critical that a proper hand off occurs so that it is clear who is responsible for what and that they have the proper training and support.

Finally, some of the more capable IIoT applications have introduced very useful advanced features and the capability to introduce custom diagnostic rules as well as apply supervised learning. Proper training on the system capabilities and how to use them enables you to get the most value out of your investment.

 
10. Ignoring Notifications and Insights

Lastly, a challenge that is not only seen with IIoT systems, and is far and away one of the most important aspects, is ignoring of system notifications and insights. There is no technology silver bullet that will transform your program results if you do not take action on the notifications (alarms, events, system support). In the same way, wearing a health tracking watch does not make you healthy, but enables you to monitor your health in order to take actions towards a healthier lifestyle. Software that facilitates task-oriented workflows, collaboration and maintenance actions will support this essential activity within the plant.

This requires that the system has the confidence of the day to day users as well as the support of Operations and Management. It is important to have a clear shared definition of next step actions and responsibilities when the system alerts. While this point seems obvious, it still remains one of the biggest challenges faced by programs today. A properly implemented IIoT system with Cloud based notifications will support taking decisions on the correct and timely actions needed to optimize operational performance.

 
References

1. “From Good To Great: How wireless and cloud technologies can advance existing Predictive Maintenance programs”, Petasense

2. Plant Engineering, “Industrial Maintenance Report”, March 2021

3. “Practical Applications of AI for Predictive Maintenance and Asset Performance Monitoring”, by Rama Redding, August 22, 2023

4. Frost & Sullivan 2022 study “Top Growth Opportunities for IoT in 2023”; “Top 10 IoT Use Cases” by Twilio’s Tobias Goebel; Tech Target survey “Top Industrial IoT Use Cases” published 2021

 
About the Author
Dan Bradley is a Mechanical Engineer with 35+ years of experience in the Reliability and Condition Monitoring industry. This includes consulting, design and implementation of programs around the world. He was previously Global Director of SKF Reliability Systems and started his career with IRD Mechanalysis, Inc.