Maciej Tabaszewski

Extraction Methods of Multi - Fault Information in Machine Condition Monitoring

An outline of the multi-fault diagnostic observation concept is pres ent d in the paper. Starting from the commonly used Singular Value Decomposition (SVD) it is shown, that it is equivalent to the modern version of Principal Component Analysis (PCA) . During the development of the concept of multi-fault observation several indices and measures obtained from SVD were tried, and some seem to bring the success. This problem is also addre sse in the paper and illustrated by real examples taken from railroad Diesel engine condition m onitoring. Introduction Advances in the metrology of mechanical and dynamic processes, and o ur growing computational ability, allow us to consider the problem of simultaneous observation of multidimensional fault space evolving in the life of operating machines and equipment. What we measure in any multidimensional observation vector, are symptoms of the machine lif e and the advancement of cumulative multi fault damage. In very many cases of machine heal th monitoring, we know from the previous records, which part of the machine and what type of symptom sh uld be monitored and classified according to some standards or critical values, comi ng from long practice. Normally for critical machines we have many critical values for each critical part, and specific direction of vibration. So, we measure horizontal and vertical vibration of the bearing caps, machine and motor casing, and some other physical symptoms, like temperature, et c. Due to this situation of multi symptom observations for multi-fault monitoring of critical machi nes, the first author has proposed in DAMAS99 [1], the application of singular valued decomposition (SVD) f or the extraction of multi-fault information from the symptom observation matrix (SOM). This matrix consist of measured symptoms as columns, and each lifetime observation gives the successive row of the matrix. Usually the symptom observation matrix (SOM) has 10 or mor e c lumns, and 20 or more rows as lifetime observations. There were several problems to solve, to enable the optimal extr ction of multi fault information from the SOM, like centering of the symptom readings, and its normal ization at the beginning. The other problem was to find the best fault indices obtained from SVD, which describe in the best way the momentary fault evolution, as well as their life (damage) a dvancement. It was shown in a several research papers that the evolution of the first singular vectors and associated singular values gives good mapping of the fault evolution. The similar problem of a choice of the best symptom from many ot hers was considered by the first author in 1980, at that time by the use of principal component analysis (PCA). So the problem arise now, maybe the PCA is the better suited for the mult idi ensional fault information extraction, in comparison to SVD method ? It was shown in this paper t heo etically, and practically on some condition monitoring cases, that both approaches are equivalent. Thi s is because in a modern computational environment, like for example Matlab , the modern PCA algorithm is basing on SVD. But using PCA gives the same results in a machine fault space, and more quickly almost on line to system monitoring. What we need, in further development of this idea, is to learn fault interpretation of SVD and PCA results, and to give them some standar dization in terms of alarms and other critical values. There is also the need to include in the proc ssing of SOM some additional knowledge on machine operation, and the behavior of symptoms (observability ) in the process of condition monitoring. It seems to be possible in the future by using g e eralized SVD/PCA concepts Key Engineering Materials Online: 2003-07-15 ISSN: 1662-9795, Vols. 245-246, pp 215-222 doi:10.4028/www.scientific.net/KEM.245-246.215

Allometric scaling and accidents at work

Allometry is the knowledge concerning relations between the features of some beings, like animals, or cities. For example, the daily energy rate is proportional to a mass of mammals rise of 3/4. This way of thinking has spread quickly from biology to many areas of research concerned with sociotechnical systems. It was revealed that the number of innovations, patents or heavy crimes rises as social interaction increases in a bigger city, while other urban indexes such as suicides decrease with social interaction. Enterprise is also a sociotechnical system, where social interaction and accidents at work take place. Therefore, do these interactions increase the number of accidents at work or, on the contrary, are they reduction-driving components? This article tries to catch such links and assess the allometric exponent between the number of accidents at work and the number of employees in an enterprise.

Similarity measures for diagnostic symptom evolution

Purpose – The observed diagnostic symptoms are often characterized by local fluctuations of their values. Hence, instead of direct observation of symptoms it is worth observing their grey models and research similarity between life curves, which can enable to guess the nature of wear. The purpose of this paper is to find useful measures of similarity of diagnostics symptoms modeled by GM(1,1). Design/methodology/approach – Measures of similarity may be used to determine the character of wear of the diagnosed object by way of comparison with known examples, which have previously been obtained and identified. A methodology for creation of such comparisons based on pre-smoothing by means of a GM(1,1) model with rolling window has been proposed. The process of smoothing enables to eliminate local fluctuations of a symptom. Their existence makes it difficult to compare symptoms. Application of a rolling window enables in turn to map the symptom properly, which may be difficult in the case of relatively short p...

Application of evolution of singular values in multisymptom diagnostics of machines

The paper presents a possibility of observation of singular values evolving in time as quantities carrying diagnostic information. To prove the usefulness of singular values for this purpose many numerical simulations have been conducted. It has been proved that when observing changes of singular values obtained from SVD during the lifetime of a machine, the appearance of reversal points must be taken into account. The appearance of such points may prove that the symptom values change abruptly (e.g. structure cracking). The appearance of such an abrupt change can easily be overlooked because of variable working parameters of the machine, which influence the values of measured symptoms and generalized symptoms after SVD. Singular values are almost insensitive to changes of working parameters, so it is easier to pick out such changes in their evolution than directly in symptoms. The paper also presents an example of application of the proposed method for real diagnostic data obtained from ball bearings.

Vibration-Based Symptoms in Condition Monitoring of a Light Rail Vehicle

Light rail systems have now their great return in many European cities carrying an increasing number of people every year. This increasing trend requires suitable operation and maintenance standards for both vehicle and track. Furthermore, in order to make a public transport competitive to private transport, its very important to increase safety and ride comfort for passengers. The aim of the presented work was to determine the suitable vibration-based symptoms for the identification of a light rail vehicle technical state, as well as the development of appropriate methodology to use the information contained therein. Both simulation and experimental phase are described. The present analysis is focused mainly on the suspension state monitoring, but some others failures were also considered.