The Kalman filter is the best way to analyze plant data with a mathematical model. If you have data that you want to understand better then the Kalman filter is a tool you should have in your toolkit. It can certainly act as a “filter” to remove noise from data, but it can do much more than that:
- It can act as a “soft sensor”, for example inferring measurements like flow rate or temperature in unknown plant streams by using heat and mass balances applied to existing data
- It can be used to compensate for lags, for example giving forward estimates of product composition or degree of conversion ahead of measurements
- It can act as a check on existing measurements, highlighting cases where measurements are suspect or likely erroneous
- It can be used to provide continuous process variable estimates between intermittent measurements
- As it is numerically robust, it can safely be deployed online to improve process control without the worry of disturbances caused by noise or erroneous measurements
Kalman Filter Structure
Kalman Filter – Dealing with Noise and Model Errors
As anyone who has ever tried to do process plant data analysis knows, noise and erroneous/missing data are severe challenges to complex mathematical model use. Subjecting most mathematical models to unprocessed plant data can be an exercise in frustration as noise is often amplified and model insights cannot be discerned. In contrast, the Kalman filter structure is designed from the beginning to deal with measurement noise. The user specifies the expected amount of noise on each measurement in the filter and these values form one key component of the filter’s estimation approach.
Similarly, any mathematical model used to analyse plant data may contain numerous approximations and limitations. Within the Kalman filter, the user specifies the degree to which each process model is to be trusted (often as a tuning parameter).
The Kalman filter then uses this information to keep the mathematical model locked on to the process data, striking the best balance between dealing with process noise and trusting the process model. As it is locked on to the process measurements, the process model in the filter can also estimate the values of unmeasured variables of interest to the user.
Chemical and Metallurgical Applications
Despite its widespread application in electrical engineering, the Kalman filter is generally under-used in chemical and metallurgical process engineering. Unfortunately there are few resources available to help chemical and metallurgical engineers build and deploy their own Kalman filters. To bridge this gap, this course will:
- Show you how to formulate a process model for use in a Kalman filter (state space approach).
- Give you specific examples of Kalman filter applications in chemical and metallurgical processes and show the advantages of the approach compared to others
- Show you how to tune Kalman filters to balance the effects of measurement noise and model errors
- Let you understand the strengths and limitations of the Kalman filter approach
- Guide you to develop your own Kalman filter for a system of relevance to you and your work