Gene Vogel
EASA Pump & Vibration Specialist

The internet is exceptionally good at one thing – providing data access. Similarly, computers which are the heart of the internet are good at dealing with data; they store data and analyze data. With the access to data provided by the internet and the storage and analysis capabilities of computers, results are possible now that could only be dreamed about in the prior century, (even if some of those dreams were nightmares).

 

While data communication, storage and analysis are the strengths of digital technology, actionable information is the goal. Achieving that goal depends on the accuracy of the data and the strength and validity of the algorithms (software) that crunch the data into information. Historically, software has been crafted that mimicked the human intellectual processes to manage data and produce actionable information. More recent software development has focused on algorithms that can manage and sort through data in a more general sense, without specific instructions on how any specific data item is handled. This category of programming is known as artificial intelligence (AI) and is intended to apply to “big data”. Big data is what accumulates from the wide array of data items related to large organizations such as:

• Corporate transaction
• Stock market activity
• Manufacturing process data
• Education; student, employee, financial
• Government
• Healthcare

A simple example of AI in the healthcare field would be the ability of a computer algorithm to identify a correlation between geographic location and the prevalence of some medical condition, without the programmer ever having identified those data items as potentially related. The algorithm is simply given diverse data sets which is searches for correlations. More sophisticated (and eerie) examples are the now ubiquitous voice activated services (Siri, Echo, Alexa, etc.). Visual recognition is a complimentary category of AI implementation.

It is important to realize that Artificial Intelligence is, in fact, artificial – it is not real intelligence. If a familiar taxonomy of intelligence is applied (Figure 1), AI is very good at Remembering, Analyzing and Evaluating; AI cannot Understand, that is a purely human attribute. In regard to Creating, computers are great at creating problems, headaches and on occasion, complete chaos. Another aspect of AI capability is the difference between correlation and cause-and-effect. AI is great at correlation, not so good at cause and effect.

Another branch of AI known as "machine learning" focuses on allowing computers to improve the accuracy of algorithm results through the use of feedback loops. If visual recognition software is given a picture of a fox, and it identifies it as a cat, it will continue to make that mistake every time it encounters a picture of a fox. But if a feedback loop is provided to “instruct” the program that it is not a cat, the next time it encounters that picture it may determine that it is a dog. If instructed that is a mistake and so on, it will eventually determine it is a picture of a fox. Hence forth it will make the correct determination – it has “learned.” Amazingly, the program will also have improved its ability to differentiate between a cat and dog in the process, even though that was not an objective it had ever been given.

An important application of AI in the realm of manufactured products is handling the Big data related to product information. For decades every new consumer product had included a registration card (or warranty registration card) to be filled out and mailed. That information allowed producers and distributors to track product location and usage, and target consumers for future marketing efforts. Warranty claims and replacement part sales augmented that effort. Today a myriad of consumer products, and industrial products, have embedded internet connections that can provide much more extensive product use data. The volume of data available can only be converted to actionable information by means of AI algorithms. But these embedded internet connections can also be a conduit for more item specific performance data, let’s call it small data. When applied in various machinery intensive applications the data becomes very valuable to the end user as well as to the producer. Prime examples are the aircraft, automotive and farm machinery industries. For more on the implication of the value of item specific performance data, Google “right to repair”.

Against that backdrop, the applications for big data and small data related to industrial machinery – the Industrial Internet of Things (IIoT) - are of special interest to EASA members. The immediate opportunities and challenges relate to new internet enabled products for monitoring the condition of industrial electric motors and related machinery. Technologies for monitoring industrial machinery have been in place for decades under buzzword titles including:

• Preventive maintenance
• Predictive maintenance
• Condition monitoring
• Reliability-centered maintenance
• Etc.

The objective of these programs is to schedule maintenance based on measurable parameters that are indicative of their condition, whether that maintenance be just a process adjustment, component replacement or repair. The benefits are the salesman’s litany; reduced maintenance costs, reduced downtime, improved safety, etc. What the internet brings to this well-established field is a new resource for managing the measurable parameter data related to the machine condition – the small data. Along with the small data are valuable Big data inputs.

A look at traditional machine condition monitoring technologies is a good place to start (Figure 2). The two predominantly monitored parameters are temperature and vibration; with rotating speed necessary for effective vibration monitoring. Classically, these are monitored with machine mounted transducers, dedicated analog signal wiring and specialized electronic instruments. For facilities with distributed control systems (DCS), a derivative system uses local transmitters to accept the analog signal and send only level information to the DCS. An alternate derivative is for a local multiplexer to accept the analog signals, provide some signal analysis and pass the results to the DCS. These permanently-wired systems also require power supplies at each instrument. Aside from the high initial installation costs, the overhead cost of maintaining these systems is also significant. The reliability of the data is another matter of concern. Suppose a critical machine were tripped due to faulty sensor or signal wire, or an instrument failure. The next time there’s a fault indicated, the first action will be to dispatch an electrician to check out the monitoring system. Any maintenance action on the indicated fault will be delayed.

Route-based monitoring is another traditional option. Portable data collector instruments are used to survey the machines for measured parameters using temporary mounted transducers. Temperature might be monitored with non-contact infrared thermometers or imagers. The biggest negative for route-based monitoring programs are the labor costs and the exposure of technicians to safety hazards when monitoring operating machinery.

The key technology enabling the new wave of machine condition monitoring systems is the availability of inexpensive, dependable wireless communication, with the internet and new battery technology playing important supporting roles (Figure 3). Wireless machine condition monitoring employs local “radio” transmitters attached to individual machines that communicate via the IEEE 802 radio frequency standard. Typical monitored parameters include:

• Operating hours or cycles
• Temperature
• Vibration
• Speed
• Motor current signature analysis
• Load (power)
  • Motors: voltage & current
  • Pumps: head & flow
  • Fans: pressure, density & volume

Details of any particular wireless method are not within the scope of this discussion. Whether by IEEE 802.11b, g, n, ac, Bluetooth or other, these are well established, dependable wireless communication protocols; the same protocols used for your local computer network.

Since both traditional and wireless monitoring systems eventually produce digital values from analog signals, the point at which the analog-to-digital conversion (ADC) occurs is a factor in data reliability. Analog signals require special wiring and a faulty signal may be difficult to recognize. Digital data is much more reliable because the communication protocols include sophisticated error checking. So the shorter the distance between the transducer and the ADC, the more reliable the system. Wireless systems use two formats for including the ADC:

1. Package the ADC with the machine-mounted transducer eliminating analog wiring and include a transmitter (radio); this device is called a sensor-node.
2. Machine mount an ADC and radio component with minimal wiring to machine-mounted transducers; we’ll refer to this as a semi-wireless system.

A very important feature of sensor-nodes is that they are battery powered, eliminating the need for power supply wiring. Sensor-nodes monitor their own battery condition along with machine condition parameters to ensure reliable data. Sensor-nodes are designed with intended battery life of 2 to 10 years, dependent on collection intervals. It is the convergence of the availability of internet communication, dependable local wireless communication and new battery power technology that make these systems viable.

Single unit sensor-nodes are simplest to apply, but the mounting location on a machine may not be optimal; on a motor the optimal mounting would be at the bearings for vibration and at the winding for temperature. Semi-wireless systems allow improved transducer location at the cost of some local, machine mounted wiring.

Sensor-nodes communicate with a gateway (aka router) which provides access to the internet via a local area network (LAN) (Figure 4). There are many possible configurations for LANs and they are beyond the scope of this discussion. Effectively, once data reaches a gateway it is available to any internet connected resource. Sensor-nodes can also communicate to the internet via cell phone networks, independent of a LAN. But for an internet resource to use the data it must be cognizant of the data format and that is controlled by the manufacturer of the wireless system. Many manufacturers have proprietary data formats. A subscription is generally required to access the data. And the data is useless without analysis, (conversion to actionable information). Open architecture systems are available which allow users to have direct access to the data. So there are two critical factors to be explored later in the discussion, the validity of the analog signal, affected by the transducer type and location, and data analysis without which there will be no actionable information. First, let’s finish the discussion of how data is communicated.

Once data has reached a gateway it is available to any internet connected resource; that includes “the Cloud” (Figure 5). The Cloud is computer jargon for massive internet connected data storage (servers) that make that data available to any internet connected device. The actual location of a server can be anywhere on earth, its location is irrelevant. It’s not necessary for the data to reside on a Cloud-based server. It could be stored on a server located locally at a facility or at least under the control of the machinery owner/organization.

In case anyone hasn’t noticed, in a very short time we have become a “smart device” oriented culture. Everyone has a cell phone and/or iPad-type device. These wireless machine condition monitoring systems work fine without smart device connectivity, but they wouldn’t be marketable. If it isn’t on a smart device it isn’t smart. Smart devices have exceptional communication capabilities – they communicate with cell networks, WiFi & Bluetooth. They can be interjected anywhere in the communication loop between the ADC and application that displays the actionable information. They are not necessary, but they make the systems attractive to the “connected” population.

Functionally, the path from data at the ADC to actionable information available to the user is irrelevant. But the quality of the data and the analysis should be a concern for any system user. All machine measured parameters are vulnerable to error at the measurement point whether the system is traditional or wireless. The concerns are specific to the type of transducer. A very skeletal summary is provided here:

Vibration – Transducer location and mounting method are critical. If a transducer is not located at the bearing housing it will miss critical data. If not properly mounted or analog signal cables are loose, false signals will result. The frequency range of the any accelerometer must accommodate likely machine fault frequencies.

Temperature – Location is critical. On an electric motor, the further from the winding the less indicative is the data of a winding fault, the same with bearing faults. Ambient temperature and air movement directly affect temperature data accuracy.

Electric motor current – A “stray flux” detector can provide frequency data but may not provide accurate amplitude data. A current transformer (CT) or Rogowski coil can provide both amplitude and frequency data, but some ADCs provide only the overall level, not the frequency data.

Speed – Many speed transducers provide reasonably accurate rotating speed, but not a phase-locked signal necessary to get vibration amplitude-phase vectors.

Technical knowledge of transducer vulnerabilities is needed when evaluating the effectiveness of any proposed machine condition monitoring system.

The concerns for accuracy of the data from machine mounted transducers are significant. The concern for the reliability of analysis results is even more so. Computer applications can provide very basic analysis information as trending and level alarms. Typically, in traditional machine condition monitoring systems an analyst, (human), takes it from there, and using software tools manually analyzes the data to determine machine faults. A number of organizations have experimented with rule based “expert” systems that analyze vibration spectral data and attempt to provide automated fault diagnosis. Most of them are right some of the time, none of them are right all of the time. To a large extent success depends on the expertise of the person(s) writing the rules (AI algorithms) and their knowledge of the specific machines. It is fair to say that at the present time, a final diagnosis of a machine fault requires that a live person with expertise be involved. Similarly, it is dubious if temperature data, motor current data or other machine load data can be adequately analyzed by AI to identify specific machine faults. Experienced maintenance professionals will use all of those parameters combined when diagnosing machine faults. An organization considering implementation of any machine condition monitoring system must consider how the necessary machine specific analysis capability will be provided.

In the discussion of "big data" it was alluded to that sophisticated AI algorithms with machine learning could be turned loose on disparate data and correlations could be established. Machinery manufacturers with access to data for a very large number of identical machines may have an opportunity to implement such an analysis strategy. At this time there is little evidence that this Big data strategy will produce accurate actionable information about machine condition. Certainly there may be some instances of success in a narrow application, but as a broad-base strategy it is uncertain what the future may hold.

From a business perspective, IIoT-based machine condition monitoring is a wide open field. As illustrated earlier, all of the individual technologies are well established; it is the convergence of technologies that has created new opportunities. There are quite a number of vendors currently in the marketplace and with off-the-shelf technologies available it is a growing field. Current vendors include:

• Motor OEMs
• Traditional vibration analysis instrument manufacturers
• Technology startups
• Machinery analysis consultants

Implementing a wireless system involves purchasing and mounting hardware, installing software applications, providing adequate LAN capability at the machine site and training users on the use and objectives of the system. It is possible a vendor could provide a turnkey system for which end users would take full operational responsibility, though that is unlikely. More likely, an end user would purchase hardware and buy a subscription from the vendor to provide some level of actionable information. That information could be specific machine fault information, abstract data analysis information, or anything in between.

Some important concerns for contracts include:

• Who owns the data?
• Who has access to the data?
• Who (what) provides the analysis?

There are also many opportunities for third party participation (EASA service centers). System vendors are interested in working with service centers. Some have dealer arrangements available. A service center may provide installation service for machine mounted components and analysis services for data, providing customers with machine condition and repair recommendations. Alternately, service centers may provide contract services to the customer on an as-need basis for technical support and analysis. Generally, at this time service centers can become involved at any level and to any degree that matches their customer’s need. It will be important to establish a relationship with a wireless system vendor that accommodates the intended customer relationship.

It is a big step between repairing machinery in a service center and being involved in a wireless machine condition monitoring system at any level. Actually, it’s quite a series of steps; there are multiple technologies involved, each with their own need for technical expertise. There is no doubt that this new technology will lower the cost of maintaining industrial machinery and help end users to realize that salesman’s litany of benefits. Some will see it as business opportunities, others will keep it at arm’s length. While it will not have a significant impact on the actual machinery repair and motor replacement market in the short term, the potential long term implications should not be ignored.

Categories: Emerging technologies/IIoT, Remote monitoring

Tags: Remote Monitoring

Rate this article:

5.0

Print