Fault detection techniques with CONDITION MONITORING
In the algorithm implementation phase of a condition monitoring based system, there are two factors to be taken into account: the first is the detection of eventual faults, the second is their diagnostics. For each of these phases there is a detection technique: Data-Driven is specific for the detection phase and Model-Based for the diagnostics.Data-based techniques (Data-Driven) are not related to the sensor, but to inputs, and basically provide a metric of similarity between the data. Those of machine learning can be an example of data-based techniques. To be implemented they require a training phase, in which the data set for healthy and defective components is defined, and a test phase or the application of machine learning techniques to the new input data. Once implemented, machine learning techniques don’t require high computational efforts and return a fast classification of the new input data. For those reasons, they are particularly suitable for cloud-computing and can be used for cloud-processing.The techniques based on models (Model-Based) can be explained through the definition proposed by Venkatasubramanian, a Columbia University professor, for which these techniques require first a knowledge of all the failures and the relationship between causes and effects. This relationship is developed using dynamic or frequency-response models. Venkatasubramanian also notes two types of models, both developed on process knowledge: qualitative and quantitative. "In quantitative models this understanding is expressed in terms of functional mathematical relationships between the inputs and outputs of the system. - he explains - On the contrary, in the equations of the qualitative model these relations are expressed in terms of functions centered on different units of the process ”.Because of the complexity and the calculation time they would require, model-based techniques are particularly suitable for off-line calculation of specific subsets of data. The results are generally better than those obtained with data-based techniques, since the description of the cause of the failure is better identified. Data-based analysis is useful for the technical development of the components, in order to optimize the geometry and maximize the expected life of the same. A criterion of choice could be the level of detail required, however the development of a model-based technique takes more time than a data-based model.It is not possible to indicate a common development methodology that could be extended to an absolute system, but it is thanks to the relevant literature that these issues can be studied in depth, thus making the choice more consistent with the system.
In the algorithm implementation phase of a condition monitoring based system, there are two factors to be taken into account: the first is the detection of eventual faults, the second is their diagnostics. For each of these phases there is a detection technique: Data-Driven is specific for the detection phase and Model-Based for the diagnostics.
Data-based techniques (Data-Driven) are not related to the sensor, but to inputs, and basically provide a metric of similarity between the data. Those of machine learning can be an example of data-based techniques. To be implemented they require a training phase, in which the data set for healthy and defective components is defined, and a test phase or the application of machine learning techniques to the new input data. Once implemented, machine learning techniques don’t require high computational efforts and return a fast classification of the new input data. For those reasons, they are particularly suitable for cloud-computing and can be used for cloud-processing.
The techniques based on models (Model-Based) can be explained through the definition proposed by Venkatasubramanian, a Columbia University professor, for which these techniques require first a knowledge of all the failures and the relationship between causes and effects. This relationship is developed using dynamic or frequency-response models. Venkatasubramanian also notes two types of models, both developed on process knowledge: qualitative and quantitative. "In quantitative models this understanding is expressed in terms of functional mathematical relationships between the inputs and outputs of the system. - he explains - On the contrary, in the equations of the qualitative model these relations are expressed in terms of functions centered on different units of the process ”.
Because of the complexity and the calculation time they would require, model-based techniques are particularly suitable for off-line calculation of specific subsets of data. The results are generally better than those obtained with data-based techniques, since the description of the cause of the failure is better identified. Data-based analysis is useful for the technical development of the components, in order to optimize the geometry and maximize the expected life of the same. A criterion of choice could be the level of detail required, however the development of a model-based technique takes more time than a data-based model.
It is not possible to indicate a common development methodology that could be extended to an absolute system, but it is thanks to the relevant literature that these issues can be studied in depth, thus making the choice more consistent with the system.