- Background The Centrifugal Pump Under Study Is Prescribed With Any Of The Following Four Conditions I Severe Blockage 1 (149.89 KiB) Viewed 28 times
Background The centrifugal pump under study is prescribed with any of the following four conditions: i. Severe blockage: 80% of the inlet is blocked ii. Mild blockage: 60% of the inlet is blocked iii. Unstable supports with mild blockage: 60% blockage and having a loose attachment of the pump to its chassis iv. Healthy: The normal state of the pump The condition of the pump is monitored using two accelerometers and a high sensitivity microphone for the various conditions above. The data will be used in the construction of a classification model with the Azure ML to identify the state of the pump based on the sensor data. The raw data from the sensors is gathered for a duration of time at varying speeds of the pump. There are 3 sensors used for monitoring: i. Two accelerometers that are perpendicular to each other. These sensors provide accelaration output in unit [g], i.e. as a multiplier of 9.806 65 ms². ii. A microphone near the pump motor. The microphone provides sound pressure output in unit [Pascal]. The data samples look like in Figures 1 and 2 below: 20 Acoustic(Pa) 10 -10 -20 severe 0 5 10 15 -50 Time(s) severe Vibration (g) 4 Time(s) Fig. 1: Examples of time domain acceleration data from the accelerometer. mild development 5 10 Time(s) 15 Acoustic(Pa) -4 mild 0 5 Time(s) 5 Time(s) Vibration (g) 10 0 10 10 development -10 0 2 2 6 4 6 Time(s) Fig. 2: Examples of time domain data of sound pressure values from the microphone.. Part 1: Pre-processing of Raw Data with MATLAB The first step is to pre-processed data from the sensors to convert it into features that will be used as the inputs for the Azure ML. Features can be constructed from the sensor data in terms of statistical quantities. In the order of significance, the top ten statistical quatities that can influence the outcomes of classification algorithms are as follows: 1. Variance of acoustic pressure values in time domain 2. Root mean square (RMS) of acoustic pressure values in frequency domain 3. RMS of acoustic pressure values in frequency domain 4. Variance of acoustic pressure values in frequency domain. 5. Kurtosis of accelaration values in time domain 6. Variance of accelaration values in time domain 7. RMS of accelaration values in time domain 8. Kurtosis of acoustic pressure values in time domain 9. RMS of accelaration values in frequency domain 10. Variance of accelaration values in frequency domain For the purpose of extracting these statistical data, a sample MATLAB code will be provided. Your task is to modify the MATLAB code to compute and output the all 10 features above. Part 2: Modeling with Azure ML Your task for the second step is to construct the classification model with the Azure ML using the available algorithms to identify the four conditions of the pump. You must consider whether using all or only selected features from the above features can results in best accuracy of classifications. You must know how to upload the features table in the .csv format you create in Part 1 to Azure. Further Learning For further understanding of the applications of condition monitoring, you can watch the following: Report and Submission Instructions 1. The Azure ML. interface does not allow you to easily download the model and visualization outputs. Therefore, you have to take screen shots of your canvas and and outputs and include them in your reports. 2. Write a report (no more than 10 pages, double space with 11 pt font) of your ML implementations and observations of the above modeling tasks. 3. Your focus of discussions should be the quality of your selected features and the classification algorithms that you use. Discuss how well these