Anyone who has ever driven a car knows from experience that some mechanical problems can be felt through touch or sound. The vibration of a steering wheel can mean that the tires need to be balanced, while a high-pitched sound coming from a brake can indicate the brake pads are worn out. A motor malfunction may be felt through the gas pedal vibration. These examples show that it is possible to validate the health of an equipment through sensoring and, in this article, we will talk a little more about industrial machinery vibration analysis as an information source for predictive maintenance.
Technologies such as automation and machine learning are being inserted into the industrial scenario at full steam. This has been contributing to businesses getting the best out of their production chains. One of the greatest highlights achieved by this technological advancement is allowing pieces of equipment to work longer without breakdowns. This is a consequence of predictive maintenance being applied to the industrial system and done with data analysis systems. Predictive maintenance comprises predicting failures in machines before they happen and avoiding non-scheduled breaks, cutting down on manufacture costs — be it with material, tools or workers. The image bellow shows where predictive maintenance fits into the maintenance engine of an industry:
Type of maintenance executed in the factory floor.
One of the main techniques globally used in the context of predictive maintenance is the extraction of machine vibration data. The information extracted from this data allows us to identify, predict and prevent failures in to rotary machines. Vibrations are an equipment’s oscillating movements around its balance position. The rotary parts of a machine generate vibrations in frequencies that allow us to identify their oscillation pattern. Any alteration in the amplitude or frequency of signals can indicate that the performance or quality of a piece of equipment is at risk. The wear and tear incurred from the mechanic friction alters the characteristics in this oscillation, enabling the detection of anomalies.
Any piece of machinery that has a rotary element can be monitored. Nowadays, the main types of equipment being put under sensoring are:
- Electric fans
- Conveyor belts
Types of Sensors
What defines the type of sensor to be applied to a piece of machinery for vibration analysis is its nature. A machine that exhibits low frequency vibrations presents a low amplitude index, an ideal situation for the use of a displacement sensor. A piece of equipment that exhibits medium frequency vibration can be monitored by a speed sensor. Machines that exhibit high frequency vibration, on the other hand, need an accelerometer to detect the high amplitude vibrations generated by their movements. The graph bellow shows the vibration lengths where each sensor works:
The intensity of the color green in the bars shows vibration lengths where each of the shows its best read results in relation to the amplitude and noise level.
The sensor most commonly used for vibration analysis is the accelerometer. Its electromechanical characteristic enables the reading of vibrations of machines and the conversion of this effect into a tension proportional to g-force (Earth’s gravitational unit of measurement). In addition, it allows data of medium and low frequency to be extracted.
The correct installation of sensors in machines is essential to guarantee quality data. Cables, connectors and mounting bases of machines must be robust enough to support temperature and humidity variation. Usually, the sensor is attached to the outside of a machine with bolts, but it can also be attached through magnetic elements or adhesives.
Types of failures
The rotary element of machines has an expiration date and this parameter is assessed through the material’s strain in the wear of rolling surfaces. However, there are cases in which the material deteriorates before its expiration date due to factors such as plastic deformities, corrosion, brine, lack of lubrication, faulty installation or even faulty projects. There are many types of failures that can be detected in rotary machines through the vibration analysis technique:
- Rolling wear
- Mechanical loosening
- Warped axis
- Electrically faulty rotor or stator
- Lack of lubrication
- Mechanical bushing gap
- Lack of alignment or balancing
- Gear tear
The sensor data analysis is usually done in the frequency spectrum. It is common to apply fast Fourier transform (FFT) to samples from the accelerometer’s sensor to verify the amplitude of the signal in the frequency interval. The amplitude of the signal in each frequency range can be taken as a report of the machine’s health. A piece of equipment working well shows indicators in known intervals. When these indicators deviate from the known range, it means that there are signs that the machine is not working properly.
The image bellow shows how this analysis can be made. In the graph bellow, data received from the accelerometer is shown in a window of time only as a visual reference of this signal. In the upper graph, an FFT of this signal is shown, representing the scale of frequency per amplitude. A few peaks from some frequency ranges jump out. Usually, the peak of highest amplitude is known as the fundamental frequency, which is where there is the highest concentration of energy caused by the machine’s vibration. The other peaks, of lower frequencies, are known as harmonics, which also indicate energy generation in more frequency ranges.
Curve of an acceleration sensor and frequency spectrum example.
The upper graph can also be read as a fingerprint or identifier of the machine’s behavior. The green columns define the limits in amplitude in which the equipment is working correctly. The red columns are amplitude ranges in which a notification or alarm should be sent to inform technicians of abnormal operation conditions.
In addition to failure detection, constant machine monitoring can be expanded, going into the concept of estimated prognosis of the future status of failure-prone components. This prognosis allows the future damage status to be defined and not only the detection of the current damage status. The damage status is defined through the calculation of the probability of failure in a function of future time use.
A robust failure-predicting system must be composed of a real-time failure detector and a future failure prognosis. Failure detection is a more mature field, thanks to great computational capacity and sensor technology to measure vibrations currently available. The prognosis is more challenging, because of the non-linear behavior of failures and uncertainties in the future prediction system, a consequence of the lack of measurements directly related to the dynamics of failures.
In this article, it is possible to understand more details of techniques of prediction through machine learning, which are the basis for failure prognosis. The failure detection and prognosis ensemble allows for the construction of the PF (Potential Failure) curve, which can be used in the planning and controlling of efficient maintenance.
The predictive diagnosis of industrial equipment failures impacts the reduction of operation costs. Extending the life of machines and guaranteeing robustness in the production chain allows new investments to be made in the factory floor.
In general terms, the investment in predictive maintenance ends up being high, since it requires qualified people and innovative systems that involve emerging technologies, such as machine learning and artificial intelligence. When it comes to managing the vibration analysis tool, the main scenarios are:
- Internal: The company buys a market solution that includes management software and sensors according to its needs. The company also hires a qualified worker to study the machines, identify critical points and locations for the installation of sensors. Training an employee to manage the tool is also an option. In these scenarios, the costs may become higher, due to system maintenance and HR costs.
- Contracted work: The company hires a contracted worker or company responsible for managing the machines. The contracted worker analyzes the industrial scenario, offers a study of critical points and budgets a monthly amount for remote machine monitoring. Currently, this is the most utilized solution.
Internalizing a predictive maintenance team and investing in pieces of equipment of collection, analysis and management of vibration data is a great viable option for bigger companies. The guarantee of great availability of the production chain directly affects competitiveness in the market and quality of the products developed. However, the situation of a smaller company is different. The current trend is to invest in contracted workers, due to the high maintenance of a team of this size to the company.
Machine vibration data extraction allows important information about the nature of an equipment to be revealed. This technique is considered the most effective in the detection of the health of pieces of equipment with rotary elements. The investment in this type of tool allows long term reductions in human resources and equipment maintenance. In addition, it also lengths the life of machines in the factory floor through predictive maintenance, an action that directly reduces the number of non-scheduled breaks.