Embedded Service Oriented Diagnostics based on Energy Consumption Data
Download as PPTX, PDF
1 like142 views
This paper presents an approach using power consumption to detect system deterioration (misalignment of conveyors) Power consumption data are correlated with workload of the conveyor system. Real time data coming from a real factory automation testbed are input to SVM for classification. The output is compared with the output of a rule-based engine.
Embedded Service Oriented Diagnostics based on Energy Consumption Data
- 1. Embedded Service Oriented Diagnostics based on Energy Consumption Data Date: September, 2012 Conference: 2012 IEEE International Linked to: eSONIA Conference on Information and Automation for Sustainability Title of the paper: Embedded Service Oriented Diagnostics based on Energy Consumption Data Authors: Corina Postelnicu, Contact information Navid Khajehzadeh, Tampere University of Technology, Jose Luis Martinez Lastra FAST Laboratory, P.O. Box 600, FIN-33101 Tampere, If you would like to receive a reprint of Finland the original paper, please contact us Email: fast@tut.fi www.tut.fi/fast Embedded Service Oriented Diagnostics 08/10/2012 1 based on Energy Consumption Data
- 2. Embedded Service Oriented Diagnostics based on Energy Consumption Data Corina Postelnicu Navid Khajehzadeh Jose L. Martinez Lastra Presenter: Bin Zhang Factory Automation Systems and Technologies Tampere University of Technology, Finland ICIAfS 2012, Beijing, China 27-29.9.2012 ARTEMIS eSONIA project (Embedded Service Oriented Monitoring, Diagnostics and Control: Towards the Asset Aware and Self Recovery Factory)
- 3. Outline 1.Introduction 2.Testbed 3.Implementation Data collection Support Vector Machine Validation 4.Failure detection model 5.Conclusions and future work Embedded Service Oriented Diagnostics 08/10/2012 3 based on Energy Consumption Data
- 4. Introduction Unexpected Financial losses failures & accidents Predictive maintenance techniques Passive: measuring data (vibration, temperature, etc), then comparing with normal values Active: injecting test signals, then monitoring responses Embedded Service Oriented Diagnostics 08/10/2012 4 based on Energy Consumption Data
- 5. Introduction Quantification: threshold settings by running the equipments until failure occurs Assumption: the measured parameters should not be influenced by other parameters Limitation: suitable for processing workstations, not transportation devices (parameters are influenced by workload) This paper associates the workload on a conveyor system to the power consumption information for failure detection. Embedded Service Oriented Diagnostics 08/10/2012 5 based on Energy Consumption Data
- 6. Testbed Embedded Service Oriented Diagnostics 08/10/2012 6 based on Energy Consumption Data
- 7. Testbed Embedded controllers to publish the device information as web services Each cell has 4 controllers Energy consumption Embedded Service Oriented Diagnostics 08/10/2012 7 based on Energy Consumption Data
- 8. Implementation: Data collection 1. Energy consumption 2. Workload Cell 5 Cell 6 Item Transfer In Item Transfer out Item Transfer In Cell 5 Cell 5 Cell 6 Embedded Service Oriented Diagnostics 08/10/2012 8 based on Energy Consumption Data
- 9. Implementation: Data collection 1. Correlation of bypass conveyor power consumption (watt) and number of pallets (0-5) 2. Power consumption of the conveyor system(watt, 1 or 2 pallets) Class 1: 0-1 pallet Class 2: 2 or more pallets Embedded Service Oriented Diagnostics 08/10/2012 9 based on Energy Consumption Data
- 10. Implementation: Support Vector Machine 1.Support Vector Machine (SVM) A classifier to provide a boundary to divide a dataset into two classes. 2.Least Square Support Vector Machine (LS- SVM) Classification is done using linear equations instead of a burdensome Quadratic equation. 70% to 80% of data are used for learning and the rest for validation. Embedded Service Oriented Diagnostics 08/10/2012 10 based on Energy Consumption Data
- 11. Implementation: Validation The 2 classes identified by the rule- The 2 classes identified by LS-SVM based engine Accuracy is computed by comparing the LS-SVM result against the rule-based engine, which shows an error percentage of 5.56% Embedded Service Oriented Diagnostics 08/10/2012 11 based on Energy Consumption Data
- 12. Failure detection model Embedded Service Oriented Diagnostics 08/10/2012 12 based on Energy Consumption Data
- 13. Conclusions and future work This paper presents an approach using power consumption to detect system deterioration (misalignment of conveyors) Power consumption data are correlated with workload of the conveyor system. Real time data coming from a real factory automation testbed are input to SVM for classification. The output is compared with the output of a rule- based engine. Embedded Service Oriented Diagnostics 08/10/2012 13 based on Energy Consumption Data
- 14. Conclusions and future work Future work Bring more parameters for analysis, i.e. vibration and temperature Embedded Service Oriented Diagnostics 08/10/2012 14 based on Energy Consumption Data
- 15. Thank you! corina.postelnicu@tut.fi navid.khajehzadeh@tuf.fi ARTEMIS eSONIA project (Embedded Service Oriented Monitoring, Diagnostics and Control: Towards the Asset Aware and Self Recovery Factory) Embedded Service Oriented Diagnostics 08/10/2012 15 based on Energy Consumption Data
Editor's Notes
- #5: Failure detection is very important in nowadays manufacturing systems, as we all know that unexpected failures can lead to delay of manufacturing process, and in turn, we get financial losses or even accidents because of that.There exists predictive maintenance techniques to prevent it from happening. One is the passive way in which we monitor data such as vibration or temperature using sensors and compare them with normal values. The other is to the active way which we inject test signals and determine the responses.
- #6: Quantification is needed in both techniques, in which we set a threshold that is the normal value mentioned before. The threshold can be determined by running the equipments until failure occurs for several times. The measured parameters should not be influenced by the environment because the threshold wont be valid.Although it may be suitable for processing workstations. But for transportation devices such as conveyors, the parameters are influenced by the workload significantly. So we need to correlate the monitored parameters with workload.In this paper, we present a method to associate the workload to power consumption for failure detection.
- #7: Thetestbed used in this paper is a multi-robot production line simulating the production of cell phones by drawing them on paper. Each cell is in charge of drawing a frame, a screen or a keyboard.
- #8: 4 embedded controllers are installed in each of the cells to publish device information as web services, for the robot, the pen feeder, the conveyors and energy consumption. The data used in this work come from the controllers for the conveyors and the energy consumption measurement.
- #9: The energy-related data are measured with a E10 energy analyser. Phase C is assigned to the conveyor system for energy measurement. The picture illustrates the wire connection. The second picture shows two conveyor systems from cell 5 and 6 working in sequence. Each conveyor system is composed of a main conveyor and a bypass conveyor. A sensor is installed in the entry point of each cell to detect if a pallet has been transferred into the cell. The transfer in signal in cell 6 indicates a transfer out of a pallet in cell 5. This way, we can determine the number of pallet in a single cell.
- #10: To determine if the power consumption is influenced by the workload (number of pallets), we performed a test on the bypass conveyor by increasing the number of pallets one by one. Figure 1 shows the result. As the number of pallets increases, the power consumption increases.Then we observed the power consumption of the entire conveyor system with 1 or 2 pallets. We can notice a significant change on the curve we obtain. So we decide to divide the workload into two categories, with class one having 0-1 pallet, and class 2, 2 or more pallets.Why there is not a class 3 (3 or more pallets)?
- #11: Classification is needed in this method so that we can locate the power consumption in a proper workload class for comparison.Classification can be done with a SVM which can provide a boundary between two classes of data.Using SVM, multi-class classification can be also achieved with a combination of binary classifications and a decision making procedure.This work is using a Least Square Support Vector Machine to classify data with linear equation. 70% of the data are used for learning and the others for validation.
- #12: The result is compared with the rule based engine result, which shows a 5.56% error on the classifier. During our experiment, we noticed a delay between the increase of workload and power consumption. This should be the main contribution to the error.
- #13: Then we developed a failure detection model with the help of the LS-SVM. First, the classifier receives the power consumption of the conveyor system and classifies the value into a class. At the same time, a rule based engine counts the number of pallets. For synchronization purpose, the rule based engine outputs the number of pallets whenever a power consumption value is sent to the classifier. Then the results from both are compared. A mismatch indicates a possible conveyor misalignment. Sychronization