Early on, condition monitoring techniques primarily involved using our senses:
As expected, in many instances, it was difficult to quantify the answers and analysis because of the dependency on the skill and experience of the observer. The observations provided powerful insight into the condition of machines but this insight was limited to the proficiency of the observer. As the use of technology for machinery condition monitoring emerged, it became easier to quantify the observations and compile historical data that could easily be used to look for trends and changes.
These hand-held devices not only allowed the analyst to quantify the condition of the machine but took some of the human element out of the observation. It allowed maintenance and production personnel to concentrate on what the quantified data actually meant:
All of the Condition Monitoring activities described above can be done online at a much faster rate, and oftentimes, more accurately than a person could perform with a hand-held monitoring device. We have gone from a snapshot of machine health every 3-4 months to a snapshot every hour or with even greater frequency, if desired. This gives us the ability to monitor many more condition parameters than before and consequently understand the true health of our equipment better. To use healthcare as an analogy, we can check the pulse of our equipment regularly, do blood tests, monitor blood pressure and even perform a detailed examination of how the whole system is working, by the click of a button — from anywhere in the world.
— Will Tudoroff, Uptime Solutions
One of the biggest challenges when collecting handheld data was the variability that the person collecting the data could create. For example, with vibration we have to go to great lengths to make sure the sensor is placed in the same place, with the same pressure, and without “contamination” — every time. Contamination means making sure the sensor isn’t rocking on an uneven surface or cleaning the magnet regularly so dirt and debris does not build up. Walking through a plant with a handheld based vibration program you will typically see pads glued or screwed to the machine or even just an “X” painted on the spot where the sensor should be placed. You have to sometimes get creative to help make sure the data is sampled the same way each time.
Oil analysis is another great example. The most successful oil programs typically have special sampling ports mounted on the machine. These programs also ensure sampling personnel have gone through extensive training to follow a very rigid process to ensure consistency of sampling and to minimize the chance of outside contamination ending up in the lubricant during sampling. Permanently mounted sensors in an online system dramatically removes the potential sources of error introduced by hand held programs. You can clean off all the paint, lightly machine the surface flat and drill and tap a stud on which a vibration sensor can be attached. The sensor then literally becomes part of the machine and ensures an accurate vibration reading is captured.
A good cleaning and machining job combined with a thin layer of a strong epoxy can also be an effective way to mount a vibration sensor to reduce error. The same principle applies to many of the new oil sensors on the market. They literally are like miniature laboratories on a chip. When mounted properly in the oil reservoir, good quality measurements of key oil parameters such as contamination, viscosity and Total Acid Number can be captured at any time. No fluid needs to be drained for the sample to be sent to a lab and allows for consistent results to be captured.
There are certain costs with online systems that have to be taken into account. When they first hit the market in the 1990’s all of the systems had to be powered and wired back to something so data could be sent. For example there were many great multi-channel vibration “data collectors on the wall” that were installed. These systems required short cable runs to the sensors which were stud or epoxy mounted to the machine. The data collection device required power and, more importantly, a communication cable such as Ethernet. Plants did not have Ethernet networks operating on the factory floor so miles of Cat 5 cable and all the network infrastructure had to be installed. These early online vibration systems provided high quality data several times a day and were a huge advancement beyond handheld based programs. However, the cost to install these systems limited their usage to only the assets that had extremely high costs of unplanned downtime.
Wireless communication plays a significant role as we dive into the condition monitoring techniques. Wireless communication has taken this high cost away from the equation, or at least dramatically reduced it. There are now numerous wireless protocols on the market that can be used in a plant setting. Some of the most popular include 900 mHz frequency protocols, WiFi and even Bluetooth. With WiFi, it’s possible to use the plant network if it exists, but most industrial wireless networks will need some simple infrastructure put in place to make it work. However, some 900 mHz wireless networks will work up to 1000 feet from a base station even with all the concrete, steel and other barriers that exist. Some systems also make their devices communication nodes almost like a repeater, to extend the range of the devices as a whole and even build in the capability to reroute data through the strongest connection, if needed.
Although new wireless protocols allow the traditional “data collectors on the wall” to communicate without running cable, they have also opened up a new style of device to the market. These are called Self-Contained Wireless Vibration Devices. They are battery operated and have built in sensors, processors, and wireless radios. Some of these devices use either single axis or triaxial MEMS style of accelerometers. They are chip based sensors that are small, low on power draw, and fairly inexpensive. Unfortunately, the quality of the data doesn’t match traditional piezoelectric sensor elements, although they can be successfully deployed on simple equipment that doesn’t have as big of a financial impact on the business when downtime occurs.
A few devices use single axis or triaxial piezoelectric sensor elements embedded in the base. These provide high quality data across a wide frequency range and rival the capabilities of traditional handheld vibration data collectors. The biggest limitation on these devices is battery life. 12-18+ months of battery life can be attained collecting data once or twice a day. If more frequent data is needed, a powered multi-channel device with wireless capabilities is better suited for the application. Some vendors supply both styles (120 V multi-channel, battery powered self-contained) of wireless devices so very efficient coverage of a plant can easily be attained by using the most cost effective solution, depending on the application and the rate of data needed. Most of the self-contained wireless vibration devices also have temperature sensors on board, that measure the bearing housing where the device is attached. One of the more interesting, effective and new transitions from handheld to online data collection, has occurred with ultrasound. Traditionally, handheld ultrasound has been used to monitor the lubrication level. High ultrasound readings indicate that a bearing needs to be greased or an oil-based system needs further analysis. With greased bearings, some handheld ultrasound units even have a built-in grease gun so the user can actually hear when the grease gets to the internal bearing components. At least one vendor supplies a wireless device that has a contact ultrasound sensor, a single axis piezoelectric vibration sensor and a temperature sensor. These sensors give the users two big benefits:
The primary purpose of Condition Monitoring software, as a condition monitoring technique, is how it helps the user turn data into actionable information. Each of the following functions of the software support the successful completion of the primary task. Data storage. The software has to be able to connect to field devices, harvest the data from them and store it in a central location. Although there are many ways do do this, key things to consider are:
Alarming. The vast quantities of data produced by an online condition monitoring system can overwhelm an analyst. Many analysts that were collecting handheld data were accustomed to looking at every piece that came in and mentally assessing it. With the increased data snapshot frequency with modern online Condition Monitoring systems that might be taking snapshots 4 times a day across hundreds of machines, this is not something that can be done. Sophisticated alarming systems are more valuable than ever. Being able to sift through hundreds or thousands of readings as they come in and pick out anomalies quickly, is crucial. The analyst literally needs a prioritized list of deviations to work from on a daily basis. Alarms that are based on statistics are very powerful, especially with the large data sets created by online systems. When combined with simple percent change or rate of change alarms, a prioritized analysis list can easily be created.
Data Analysis. Once the alarming system creates the prioritized list then the skills of the analyst take over. The software must have a powerful set of data analysis tools so the analyst can effectively and efficiently diagnose problems. This is where the data is turned into actionable information. The analyst will dive into the data and determine the following:
To do this successfully, the diagnostic tools in the software must allow the analyst to look at the machine holistically. All data collected on the machine whether it be process, vibration, temperature, ultrasound or oil quality, must be viewed at once to create the true picture of machine health. The software has to be able to handle the high-density vibration waveforms and spectrums can be manipulated, processed and displayed as necessary to determine the exact failure modes. The trends can be used to see how all these parameters are changing over time and assess the progression of these modes so severity can be accurately assessed.
“The essential aim of Big Data is to gather a large enough training library of fault information that the software would be able to match a known fault in the library with an emerging fault in a subject machine.”
Big Data is an industry term for gaining value from correlating massive amounts of disparate data from many sources, to find insights into business and reliability problems.
The systems that gain insights from Big Data are based around the need for more and faster data. However, these systems have struggled when it comes to handling condition monitoring data, particularly vibration data. When you look at the history of successful condition monitoring programs, many were built on monthly handheld data collection. In most ways, more data has had an incremental effect on the success of condition monitoring programs. If anything, online systems have improved the quality of the data with more frequent data collection being an added, not primary, benefit. Thus, many online condition monitoring systems were never designed with Big Data in mind, since collecting data more than 4 times a day rarely has any benefit to analysts.
The second barrier to pushing condition monitoring data into the Big Data arena specifically, has to do with vibration readings. Vibration readings are extremely complex and dense. A typical vibration reading is taken by sampling the movement of the machine at a rate of 20,000 times per second. That high sample rate allows the analyst to observe exactly how the machine is vibrating. Only a few seconds data is needed to do this. The dense snapshot might be made up of 100,000 data points that have to be kept together in order to be useful. Each individual sample is meaningless. The whole snapshot of samples is needed to manipulate, visualize and analyze. An analyst literally treats that dense, 100,000 data point sample, as a single measurement point-in-time.
Most Big Data systems struggle to handle this type of data. While by definition, Big Data needs more volume to be effective, this dense vibration data snapshot is effectively only a single measurement. These systems are built to handle hundreds of thousands of individual measurements, not a single measurement made up of 100,000 data points.
Many companies have invested in enterprise or plant wide systems to collect, synthesize data, and help with certain processes and procedures. Maintenance Managers and Reliability Engineers regularly use Computerized Maintenance Management Systems (CMMS) and Plant Historians. CMMS systems are tools to help the maintenance function be more effective and efficient. They help maintenance plan, organize, schedule and track all maintenance activities. A CMMS system is arguably the most important tool the maintenance department has at its disposal. Some companies will want their condition monitoring software integrated into the CMMS system. Although many CMMS systems have easy ways to accept scalar data and alarm indications from Condition Monitoring systems, this type of integration provides very little value. Being able to trend a vibration or temperature level in a CMMS or have a Good or Bad indication of machine condition is really limiting. The best integrations between Condition Monitoring systems and CMMS involve actionable information.
The whole goal of Condition Monitoring is to determine exactly what is wrong with the machine, how far along the failure progression curve the machine is, and the best course of action to address the problem. We call this actionable information. It is much more valuable to take the problem diagnosis and suggested resolution, and pass this to the CMMS. That way a work order can be created and prioritized properly. Having an integration that indicates: “high vibration of .362 in/s” is almost useless when compared to sending: “high vibration due to third stage bearing wear. Failure will likely happen in 2-3 weeks. Recommend bearings be replaced as soon as possible.”
Plant historians are repositories for all the process data that the control system collects and uses during operation. Most plants have very sophisticated control systems and maintain large historical databases of all the parameters collected. These historians can handle scalar condition monitoring data but not detailed vibration data such as waveforms and spectrums, that are used when a full analysis and diagnosis of a problem is warranted. However, there is value in storing some condition monitoring data in the plant historian, along with all the process data. Some customs want as many different data sources as they can populate the historian. Usually customers are looking for correlation between different sensor data collected from a given machine or process. A bi-directional connection between the Plant Historian and the Condition Monitoring system is preferable. Not only can condition monitoring data flow to a single repository, but process data that can aid in understanding the true condition of the machine can go the other way.
The most sophisticated condition monitoring systems have multi-dimensional alarming capability which shows the correlation between vibration data and process data. Things like machine speed, product being made, pressures, feed rates and flow rates can noticeably affect the vibration data. Good analysts can mentally account for the different operating conditions when looking at data. Excellent analysts use tools in the software like multi-dimensional alarms to give accurate diagnosis.
Now the industry has evolved even further, with an advanced solution set — a turnkey wireless condition monitoring solution. Such end-to-end solutions take the customer from design to data integration.