Skip to main content


Production machines are becoming increasingly intelligent. They contain more and more sensors to monitor their condition and operation and are often connected in a Cloud environment. This allows to monitor their condition and operation continuously and from a remote location, to compare findings about several similar machines and to adjust their operation.

Figure 1: Local bearing fault caused by surface fatiguing on account of the bearing ball trajectory

Manufacturing companies continuously try to increase their productivity, among others by avoiding machine downtimes. The latter bring about considerable costs because of the resulting loss of turnover. Early fault detection in, for instance, bearings (figure1) and gears enables to realise a serious cost reduction and to gain time.

However, up to his day, the market does not yet offer suited solutions for this. All too often, machine operators must still base their maintenance decisions on data from the periodical and manual inspection of one single machine, which does not always result in correct conclusions. Furthermore, the closed format of commercial monitoring systems does not allow to improve specific applications in a custom-made manner and the existing solutions are too expensive.


In the project "Cost-Effective Vibro-Accoustic Monitoring", Flanders Make and its project partners developed algorithms for using inexpensive sensors for permanent monitoring, which are also adjustable to different industrial applications.

The monitoring system consists of smart algorithms in combination with sensors and an open embedded platform. This offers opportunities for permanent condition monitoring applications. This permanent monitoring enables manufacturing companies to increase the quality of their products, lower their production costs and save time.

Figure 2: Building blocks of monitoring system.

The developed monitoring system is mainly intended for use on rotating machinery, particularly on bearings as these are high-maintenance components that are crucial for the correct operation of machines.


To enable an effective fault diagnosis, it is important to get insight in the failure behaviour of the relevant machine components. The number of failure mechanisms in bearings is very diverse: from surface fatiguing up to lubrication-related failure mechanisms. 

A significant part of the bearing faults arises very locally and result in impulsive vibration behaviour. Other vibration sources in machines such as gears, shafts and noise typically conceal the effects of bearing faults. 

Figure 3: Vibrations in machines are predominantly caused by bearings, gears and shafts.

So as to isolate the aimed at bearing fault information from the vibration signal, a so-called smart algorithm is used. This algorithm is composed of three signal processing steps.

Vibrations caused by gears are often the dominant factor in the vibration signal of rotating machinery. These vibrations are removed in the first step. 
In a second step, the noise in the filtered vibration signal is dampened. In the final step, the impulsive vibration behaviour caused by a bearing fault is strengthened.

Figure 4: The three predominant signal processing steps in the algorithm: 1) removal of dominant vibrations caused by a/o gears, 2) reduction of noise, 3) strengthening of impulsive vibration behaviour caused by a bearing fault.

The presence of bearing faults is then detected using the ‘squared envelope' frequency spectrum. The amplitude of the specific frequencies of the bearing fault is used as an indicator. Subsequently, the calculated feature is compared with a pre-defined threshold value. A healthy condition is reported if the value is smaller than the threshold value whereas a defective condition is reported if the value exceeds the threshold value.

The entire signal processing and analysis process is performed automatically in an embedded platform without any expert intervention being required. The required information is limited to the shaft rotation speed and a specification of the type of bearings in the relevant machine.


The high investment cost is one of the bottlenecks for the wide application of condition-based maintenance strategies. A significant part of these costs can be ascribed to the sensors. Sensors based on micromechanical systems (MEMS) offer an inexpensive alternative to more expensive high-end sensors. MEMS-accelerometers combine economic advantage with a compact design, high sensitivity and temperature stability.

However, these MEMS-accelerometers are not yet generally used in the industry for condition monitoring purposes. This is partly due to the fact that they do not always meet the high requirements in terms of bandwidth, dynamic range or low noise level. 
We selected the MEMS-accelerometer ADXL1002-50 from Analog Devices and the piezo-film accelerometer ACH-01 from Measurement Specialties. 


Figure 5: Market overview of inexpensive analogue accelerometers. The dotted lines indicate the minimum condition monitoring requirements. The diameter of the circles represent the level of noise as specified by the manufacturers.


We also developed an embedded platform to measure and process machine vibrations. The main design criteria were compactness, openness, scalability and price.

The solution that we developed consists of a Beaglebone Black single-board computer equipped with a Linux control system and a custom-made interface. The system has 6 analogue input channels that can be sampled at 52 kHz. It can be programmed in open-source programming language Python.


Figure 6: Top: MEMS-accelerometer integrated in a custom-made enclosure. Below: Open and embedded development platform equipped with 6 analogue input channels.


Analyses have shown that the inexpensive sensors and embedded platform perform equivalently as high-end alternatives. The performance and reliability of the algorithm has been extensively tested on several types of bearings and bearing faults. The research project has shown that the algorithm is able to detect smaller faults (ø195 µm) than the analysis methods as these are traditionally used in the industry (Figure 7). This allows to detect faults in an early stage. Besides, the algorithm has been tested on a wide range of applications, from simple to more complex machines such as, for instance, gearboxes.

Figure 7: The Flanders Make algorithm detected smaller bearing faults than traditional bearing monitoring methods.


The research partnership between academic and industrial partners has enabled the early detection of small faults using smart algorithms and inexpensive open hardware. 

Furthermore, we now have the necessary building blocks at our disposal to adjust the monitoring solution to specific industrial applications. This allows for a wide and permanent application of the monitoring system on rotating machinery such as gearboxes, pumps, motors, compressors and production machines

Figure 8: A graphic interface that is used to control the monitoring system and show the diagnosis results of the algorithm.


Thanks to its scalability and openness, the developed condition monitoring system can be integrated into a 'fleet' environment so that a group of similar machines can be monitored simultaneously. 

This full-fledged digitisation (Industry 4.0 and Cloud-connectivity) allows to gather data on machine faults of similar machines. This will improve the quality and reliability of maintenance decisions and consecutive actions, resulting in a significant reduction of unwanted downtimes. In addition, the solution also supports important decisions related to the performance management of machines (e.g. energy efficiency, quality of manufactured product). As a result, machine operators will be much better supported when defining and executing optimal machine settings.  This opens new possibilities for machine builders to explore new business models based on digital services, selling not only physical products but also digital services for optimal machine maintenance and performance management.


This research has been conducted within the scope of the Vibmon project by the Flanders Make DecisionS corelab in cooperation with the academic partners Antwerp University (CoSys-Lab) and KU Leuven (PMA). The project receives financial support from VLAIO.