Predictive maintenance of pumps based on signal analysis of pressure and differential pressure (flow) measurements

Abstract

Maintenance of process equipment is one of the inescapable tasks associated with the operation of process plants and, until relatively recently, it was implemented either on a routine basis or after the failure of equipment. Attitudes are changing and now many organizations are adopting methods for identifying incipient faults, so that maintenance can be scheduled before there has been a failure which would lead to loss of production and spoilage of raw material. These methods include schemes in which all the measurement signals in a plant are gathered and, in addition to providing the input to the various control systems, they are compared with model-based values for individual sections throughout the entire process, so that a warning or an alarm can be raised, according to the seriousness of the observed discrepancy. Although these methods represent a considerable advance on ‘routine maintenance’ and ‘repair after failure’, their effectiveness is limited by the quality and detail of the information that is gathered. Each process measurement should be regarded as a ‘window’ through which the operation of the process can be observed. However, the majority of existing process measurement systems have been designed to provide the input signal to a process control system, for which the primary requirement is a good average value. Typically, this is obtained by filtering out all the components of the sensor signal having frequencies greater than about 5 Hz, with the result that the ‘window’ is only translucent, rather than transparent, and much of the detailed information which is available at the interface between the process and the sensor is lost. If, on the other hand, the measurement system has a wide frequency response, then much of the information that is available at the interface between the process and the sensor can be gathered and analysed to provide a clear view of the operation and status of both the process and the measurement system. This concept involves two significant departures from the conventional approach to process measurements and to the utilization of measurement information. In the first place, the measurement systems should be designed to have the widest possible frequency response, so that the maximum information regarding the measured process parameter that is present at the interface between the process and the sensor is gathered. Secondly, instead of conditioning the primary sensor signal to provide a good average value for process control, it is analysed using well established techniques to provide not only the signal for process control but also information on which predictive maintenance procedures can be based.

Mathematical Modelling of the Centrifugal Compressor

“…GMDH is used to synthesis the mathematical model of a centrifugal compressor (5). The classical GMDH has its disadvantages: the possibility of losing some or all of informative parameters, if they are excluded at the beginning of a selection; output vectors of the best models become more correlated due to an increase in the amount of series.…”

“…Scheme of the information transfer about the compressor technical state Therefore, thermodynamics and outlet parameters can be used for creating the centrifugal compressor mathematical model including all basic requirements (the adequacy, the model universality, the possibility of the model implementation) [2]. The influence of random and predictable processes (wear, external and exploitation conditions, etc) must be taken into consideration during the creating the compressor mathematical model [3][4][5][6].…”

Design and Implementation of a Fuzzy Classifier for FDI Applied to Industrial Machinery

“…Examples of industrial machineries which attracted the attention of the FDI research in the last decades are machineries in industrial processes, manufacturing processes and theoretical processes ([ 25 , 26 , 27 , 28 , 29 ]), e.g., motors ([ 30 , 31 , 32 , 33 , 34 , 35 , 36 , 37 , 38 , 39 ]) and pumps ([ 40 , 41 , 42 , 43 , 44 , 45 , 46 , 47 ]). In [ 25 ], PCA is exploited together with the expectation formulas of T 2 and squared prediction error statistics in order to achieve the detectability conditions of different faults in a simulated double-effective evaporator process.…”

“…With regard to pumps, model-based fault detection for centrifugal pumps and AC drives is proposed in [ 40 ]; the role of parameters’ estimation in the diagnosis of pumps is investigated. A predictive maintenance procedure is developed in [ 41 ] based on signal analysis of pressure and differential pressure flow measurements. In [ 42 ], the design and initial testing of a quantitative model-based FDI scheme for a water pumping system is proposed.…”

Dynamic analysis and fault diagnosis of a water hydraulic motor

“…Higham & Perovic [58] studied the predictive maintenance of pumps based on the pressure and flow signal analysis. Neural network is the most important method in the fault diagnosis of mechanical system.…”