Acquisition and processing of new data sources for improved condition monitoring of mechanical systems
This research is focussed on the development of new ways to perform on-line condition monitoring in mechanical systems. It is focused on technologies that have been quite scarcely investigated in this field, in particular, the use of lubricant oil viscosity monitoring and motor current signature analysis technologies for improving the condition monitoring in gearboxes. On the one hand, a new solution based on magnetoelastic materials is presented for the monitorization of lubricant oil viscosity. On the other hand, motor current signature analysis (MCSA) is presented as a counterpart to traditional accelerometers for the monitorization of mechanical anomalies.
Broadly speaking, nowadays the most common maintenance practice for gearboxes is preventive maintenance. Lubricants and different mechanical parts such as gears and bearings are replaced on a periodic basis regardless of their health state. Steadily, condition based maintenance is gaining importance, as it offers several advantages over preventive maintenance. Still, one of the drawbacks of condition-based maintenance is the need to assess the health state of the component, either by adding sensors (on-line) or by making punctual measurements by qualified staff (periodic inspections). As the price of sensor technologies tends to go down and more electronics is incorporated to the machinery, the trend is to take advantage of the existing on-line monitoring options.
Lubricant oil is similar to the blood that flows inside us. Problems in our body have their reflection in blood, and in an analogous manner, problems in the gearbox will be manifested in the lubricant oil. That is the reason why lubricant oil must be monitored., and among the main lubricant properties, viscosity is the most important one. If oil viscosity does not remain in the right interval, lubrication won’t accomplish its purpose, risking even a catastrophic failure of the mechanical component. However, viscosity measurements are still made via off-line methods, using slow and costly equipment systems. Therefore, in this Thesis the concept of on-line lubricant oil viscosity condition measurement is examined.
In particular, a magnetoelastic kinematic viscosity sensor for on-line or in-line measurements is designed, built and tested. The main advantage of the sensor prototype proposed is its capability to measure oil viscosity in a wide range of values (from 32 cSt up to 320 cSt), which is not known to any other sensor commercially available.
Magnetoelastic materials exhibit an intimate coupling between their magnetic and elastic properties, in such a way that a magnetic excitation produces an elastic response of the material and vice versa. The principles governing the magnetoelastic effects and their beneficial use for the design of the on-line magnetoelastic viscosity sensor are explained in the Thesis.
Two different prototypes are presented. The first one is intended as a proof-of-concept of the technology, and the second as a prototype of a practical device for the measurement of the viscosity of different oils. The signal processing to correlate the magnetoelastic response with the viscosity of the oils is described, including the use of a new phenomenological model. Likewise, the relationship between the temperature and the measurements has been studied.
Concerning motor current signature analysis (MCSA), the objective of the Thesis is to advance in the design of a system that can monitor a gearbox in normal operation. In this sense, the work is oriented towards the analysis of transients in speed, maintaining the load fixed. A gearbox test bench is used to reproduce different faults and acquire data in different operating conditions. With respect to the analysis of the current signals from the motor moving the gearbox, wavelet analysis is selected as the most convenient technique for the analysis of transients in speed.
The Thesis describes the experimental part of the work, including the test bench, the design of experiments, the type of gearbox faults monitored, and the organization of the data pool. The procedures for data reduction, preprocessing and analysis are presented, which produce different sets of features describing the health state of the system. The performance of these features is assessed using different classification algorithms and the results are discussed and compared including the comparison with other pre-processing techniques, such as dual level time synchronous averaging.
The techniques developed have been applied to both gears and bearings inside the gearbox.
The investigation performed demonstrates that the combination of using transient information from the feeding current and the use of wavelets to analyze the data, maximize the value of motor current signal for condition monitoring in a gearbox, and enables its widespread deployment in maintenance procedures.