After I encountered situations that conflicted with my ethical understanding in the condition monitoring and predictive maintenance sector over and over again, I decided to write this article for informational purposes for those who are considering investing in these fields. Although this is an article based on my personal experiences, if anyone gets offended, my apologies in advance.  

(I know that this is a long article. If you wish, you can jump straight to the end and read the conclusion.)

Due to the recent pandemic, the increasing need for digitalization also manifests itself in the field of production. While production should not come to a halt, human dependent processes lead companies to condition monitoring systems. With the increasing demand, the number of products supplied to the market is also increasing each passing day. Because vibration, which is the most important parameter of condition monitoring systems, requires expertise on its own, it is becoming difficult to evaluate products by maintenance engineers. This situation causes the engineers to be prone to be misled by the salespeople of the companies. Since I am the owner and chief designer of a company that develops products in this field, I regretfully see that I am obliged to raise the awareness of customers objectively as much as I can.

Let’s review the products under two main headings:
✓ Hardware
✓ Software


Vibration, temperature, current/voltage, pressure, ultrasonic/acoustic measurements are the types of sensors found in products. The most common sensors to be found are vibration and temperature sensors. You will probably see the temperature sensor in all products since the temperature sensor is added to every product as it is both cheap and easy to use. The device measures its internal temperature, the value read does not give accurate data of the measured device, yet it is satisfying in most instances because it is correlated. The main focus will be on vibration.

The rest of my article will be about vibration sensors. Since other sensor types are not my areas of expertise, it would not be right for me to put my two cents in. I can state that even though the sensor types are different, “Data Acquisition” techniques will not differ and the same rules can be adapted to different data types.

Before going into the details of vibration sensors, I want to make the following warning: It’s almost 2021, please don’t invest in sensors that only provide RMS, CREST, Kurtosis, etc.

You will find vibration sensors as accelerometers in this field (Vibration can be measured as acceleration, velocity, and displacement; but I will not go into the details because, in the condition monitoring field, all of these values are measured as acceleration and converted into other units by calculations). You will see accelerometers in basically two types, MEMS and Piezo. You don’t have to make a choice by looking at these types. Because this difference will affect some features (I will talk about them in the rest of the article), it is sufficient to look at the result, not the technique used.

Frequency Range: This should be the most important factor in your choice. Of course, if your operating temperature is 150 degrees Celsius, you cannot use a sensor that works at a maximum of 100 degrees Celsius. I exclude such cases.

This value is where companies deceive customers the most. Ask this clearly: what are the sampling rate and linearity values separately? Unfortunately, I witness this in the market; the frequency range is given as 4 kHz for example, yet the sensor is linear up to 1kHz and the sampling rate is 4kHz. In this case, the users will only see the vibrations up to 1kHz in their measurements correctly. There is also the following situation; ask whether these properties are the same for each axis in three-axis sensors or not. You may come across this in the technical documents:

(You may not read the technical detail.)

Technical detail: The concept of correct measurement that I mentioned in the concept of linearity represents a range in vibration measurements called -3dB. Just as no measurement type is one hundred percent accurate, in the vibration field, the measurement is considered accurate until there is half the difference in the frequency spectrum between the measured value and the actual value. In some sensors, this linearity is given separately as 5% and -3dB (Usually when used in the R&D field). I respect that.

Frequency Resolution: This factor is a calculated value. The ratio of the sampling rate to the number of sampling gives this value, and it is related to the controller and the memory unit controlling the sensor. The longer you take measurements; the higher resolution spectrum you will get. Nevertheless, this doesn’t mean that it will always give good results. It may cause you to collect different signals that mask the error signal and cause the error signal to appear with lower intensity in the spectrum. I won’t continue, but those who are concerned can search it as “Fourier Transformation” and “Nyquist Theory / Sampling Theory”.

Measuring Range: This value represents the maximum acceleration value you can measure. Don’t look at this as the higher the value, the better; since as the range increases, the next factor, the noise intensity, also increases. The best choice for this value would be the lowest possible setting. It is substantial that the sensor is adjustable, thereby it can be used at different spots. The highest acceleration value I have ever come upon was around + -14g in a semi-hermetic compressor. We promptly directed the company to take it under maintenance. You can act like this; while your system is running, take measurements at the spots to be placed and choose a sensor that can measure 1.5-2 times its maximum.

Noise Intensity: This value is frankly a quite technical situation; naturally, you can’t comment on this as a maintenance engineer. Let me explain how this unit affects you. The higher the noise in the measurement, the harder it is to capture the error signal. You can think of cracks in large mechanisms such as bearings and gearboxes as the error signal. These cracks create beats of small amplitudes and as the noise increases, these beats are hidden behind the noise. The larger the crack, the larger the amplitude, making the beats distinguishable from noise. In other words, the lower the noise, the sooner the chance to detect the error.

Data Transfer: It is not quite possible to comment on data transfer as good or bad. I cannot make any other guidance other than giving a rough answer like the right one should be decided according to the area of use. At the end of the day, we want to store the measurement data somewhere and then analyze it. At the start, this data needs to be transmitted from sensors to storage sources. The communication model in between can differ from product to product. The necessity for continuous cohesion of IoT products also creates a requirement for infrastructure. The capacities of the ordinary network (modem) devices will remain incapable to meet the need, pay attention to this issue.


I begin by dividing the software part of the article into two parts:

Data Analysis

Data Analysis: Some standard statistical calculations are made after the vibration data is taken. Please request this data. We even offer it as open-source, you can ask your manufacturer to add it. Ahead of statistical methods, artificial intelligence and machine learning are also a quite popular part. We frequently hear these topics on the predictive maintenance side as well. In fact, lots of products offered under the name of artificial intelligence give results based on ISO10816 and similar standards. Products that claim to achieve results with marvelous and wondrous algorithms cannot go beyond to reduce noise with traditional signal processing methods. I am not saying that they do not work and are useless, but when you come across such products, ask for a demo with your own equipment in the field. Be cautious with manufacturers who create a demo with error signals generated in Matlab and claim that it works.

Architecture: The last part of the article is about software architecture, you will experience the problems in this field when you purchase the product and come to the integration stage. Before talking about the issues that need attention and consideration, I would like to point out to the maintenance engineers who read my article; there is a high chance that the IT staff in your company does not have the required proficiency for the integration process. When choosing a product, choose a company that you can get integration support from.

On the software side, solutions will appear in two ways. The first one is called On-Premise, which will run enclosed in your network infrastructure, and the other one is the software operating in the internet environment, called the Cloud. The only advantage of On-Premise solutions is that if you are concerned about data privacy, it’s the less worrying option on the security side since it doesn’t go to the internet. I think this situation will decrease over time because economic reasons make the prices of Cloud solutions uncompetitive against On-Premise solutions. Cloud systems are becoming advantageous in every aspect for the manufacturer, and therefore, day by day, they can be offered to customers at more advantageous prices. The sole thing that needs to be done for integration in Cloud systems is to allow the devices to access the internet. On the other hand, the needs are many in On-Premise. It should not be approached with the thought that Cloud systems cannot be connected to your internal system. Speaking in terms of our products, our system allows inputs and outputs such as raw and processed data output, measurement command controls, and we provide free support of their integration. Ask detailed questions about these issues for the products you are considering purchasing because you can make an investment and then pay much more for integration services for functionality.


If you are the person to make the purchase decision, read the full article.

Never invest in sensors that do not provide spectrum and give only the RMS value. Laugh away those who charge extra fees.

Test the algorithm/artificial intelligence products, which give marvelous results, in your own facility with a demo.

Do not insist that the whole system works in the SCADA system, request the input and output integrations needed.

Work with companies that you can get support for integration.