Signal Processing Technique Which Separates Signal Components in a Sensor for Sensor Diagnostics

The present invention relates to process variable sensors of the type used in the process control industry. More specifically, the invention relates to life expectancy estimation and diagnostics for such a process variable sensor.

Process control transmitters are used to monitor process variables in industrial processes. For example, a transmitter might monitor pressure, temperature or flow (e.g., process variables) and transmit such process variables back to a control room, where a controller sends a control signal back to an actuator (e.g., valve, motor) to control the process. In order to monitor a process variable, the transmitter must include some type of a sensor. For example, transmitters include sensors with resistances or capacitances which vary in response to temperature, deformations or strain which allow the transmitter to measure, for example, temperature, pressure, flow, level, pH or turbidity.

As sensors age or are subjected to harsh environmental conditions, the accuracy of the sensor tends to degrade. Ultimately, the sensor will fail. Diagnostics can be performed on a sensor by monitoring the sensor output signal. For example, a simple diagnostic technique is to compare the sensor output to a maximum or minimum value and provide an alarm indication if the threshold is exceeded. However, one difficulty in prior art diagnostic techniques is that the variations in the process variable being sensed should not be incorrectly interpreted as a sensor fault.

A device in a process control system includes a sensor input which receives a composite sensor signal from a process variable sensor. The composite sensor signal includes a process variable signal related to the process variable being sensed and a residual sensor signal related to sensor operation. Wavelet preprocessing circuitry coupled to the sensor input separates components of the composite sensor signal and responsively provides the components of the sensor signal to diagnostic circuitry. Diagnostic circuitry receives and responsively provides an output related to sensor health.

FIG. 1 shows a process control system including a transmitter in accordance with the present invention.

FIG. 2 is a simplified block diagram of diagnostic circuitry in accordance with the present invention.

FIG. 3 is a simplified block diagram of a process device in accordance with the present invention.

FIG. 4 is a graph showing a nominal base signature for a sensor signal.

FIG. 5 is a diagram illustrating an individual wavelet transformation.

FIG. 6 is a graph illustrating various components of a process variable sensor output from a wavelet decomposition of a composite sensor signal.

FIG. 1 is a diagram of process control system 2 including field mounted transmitter 40 coupled to control room 4 over a two wire process control loop 6. Transmitter 40 monitors a process variable (e.g., flow, pressure temperature) of process fluid in process piping 8. Transmitter 40 transmits information related to the sensed process variable to control room 4 over loop 6 by controlling the current flowing through loop 6. For example, the current flowing through loop 6 may be controlled between 4 and 20 mA and properly calibrated to indicate the process variable. Additionally or in the alternative, transmitter 40 may transmit digital information related to the sensed process variable over loop 6 to control room 4 such as in a HART.RTM. or Fieldbus protocol. Transmitter 40 includes circuitry described herein in more detail which provides advanced diagnostics including life expectancy information (e.g. health) related to sensor operation. The present invention may be implemented in, for example, a pressure transmitter, a magnetic flowmeter, a coriolis flowmeter, a level transmitter with a low power radar measuring means, a resistance based temperature transmitter, or any other type of transmitter.

FIG. 2 is a simplified block diagram of one example of circuitry for performing sensor diagnostics in accordance with the invention. As shown in FIG. 2, process variable sensor 20 and sensor compensation circuit 22 provide a composite sensor signal to measurement circuitry 11 and preprocessing function 14. Measurement circuitry 11 provides an output to output circuitry 13 representative of the process variable being measured. The composite sensor signal provided by sensor 20 includes a component related to the sensed process variable and a residual sensor signal due to mechanical or electrical characteristics (e.g., the transfer function, process noise, etc.) of sensor 20. Further, the signal related to the process variable may be separated into two components, one component due to repeatable sensor variations and another due to repeatable process variations.

Wavelet preprocessing function 14 receives the composite sensor signal and separates the individual signal components, including seasonal variations. These separate signals are provided to diagnostic circuitry 12 on data bus 15. As explained below, this allows diagnostic circuitry 12 to function on the separate, individual signals which make up the composite sensor signal, and, for example, provide a diagnostic output indicative of a failure of sensor 20.

Output circuitry 13 receives the process variable from measurement circuitry 11 and formats the output as desired. For example, the output may be coupled to loop 6 shown in FIG. 1. Further, output circuitry 13 receives the diagnostic signal from circuitry 12 which, for example, may be output on loop 6, used to inhibit the output of process variable or to alarm.

FIG. 3 is a simplified block diagram of transmitter 40 in accordance with the present invention including wavelet preprocessing circuitry 32. Transmitter 40 includes sensor 20 which provides a sensor signal to sensor circuitry 22. Sensor 20 can be a resistance based sensor for sensing pressure or temperature (e.g., an RTD or a strain gauge), a capacitive pressure sensor, etc. As discussed, the sensor signal is a composite signal which includes a process variable signal and a residual sensor signal. Sensor circuitry 22 performs initial compensation including optional scaling, on the analog sensor signal and the output of sensor circuitry 22 is converted into a digital format by analog to digital converter 24. A microprocessor 26 receives the digitized process signal and is also coupled to memory 28 and a system clock 30. Microprocessor 26 operates in accordance with instructions stored in a memory 28 to perform various functions. Two such functions in accordance with the present invention are shown as blocks within microprocessor block 26. Specifically, microprocessor 26 includes substantially all of wavelet preprocessing circuitry 32 and sensor life expectancy and diagnostic circuitry 34. Outputs from microprocessor 26 are connected to input/output circuitry 36 and coupled to process control loop 6. I/O circuitry 36 also provides a regulated voltage output which, in some preferred embodiments, powers all of the circuitry of transmitter 40 received through process control loop 6.

Prior art diagnostic circuitry is often unable to separate the signal due to the process from the signal arising from the sensor and its transfer function. Thus, the prior art frequently cannot distinguish whether a recognized problem is caused by the sensor or the process. In contrast, the wavelet processing circuitry 32 of the present invention separates the composite sensor signal into its separate components. The separated sensor signals are provided to life expectancy and diagnostic circuitry 34 which is thereby able to more accurately determine life expectancy and diagnose sensor operation. Circuitry 34 receives a signal in which the "noise" from the process has been substantially eliminated. Circuits 32 and 34 may be realized in analog circuits, separate digital circuits, or through a microprocessor 26 as illustrated in FIG. 3.

Microprocessor 26 provides a signal related to the process variable and a life expectancy and diagnostics signal to I/O circuitry 36. I/O circuitry 36 transmits information over two wire loop 6 in accordance with known techniques as in a fully digital protocol such as Fieldbus or WorldFIP, a hybrid analog/digital protocol such as a 4-20 mA signal with a digital signal superimposed (e.g., HART.RTM.), or even according to the DE protocol. Further, microprocessor 26 may receive instructions sent from a hand-held communicator or control room 4 over loop 6.

The present invention can also be practiced in software resident in any of a number of places in a process control system such as in a field mounted controller, a remote PC or controller or even a final control element such as a valve, motor or switch. Furthermore, modern digital protocols such as Fieldbus, Profibus and others allow for the software which practices the present invention to be communicated between elements in a process control system, and also provide for process variables to be sensed in one transmitter and then sent to the software in a different piece of equipment.

Wavelet analysis is a technique for transforming a time domain signal into the frequency domain which, like a Fourier transformation, allows the frequency components to be identified. However, unlike a Fourier transformation, in a wavelet transformation the output includes information related to time. This may be expressed in the form of a three dimensional graph with time shown on one axis, frequency on a second axis and signal amplitude on a third axis. A discussion of wavelet analysis is given in On-Line Tool Condition Monitoring System With Wavelet Fuzzy Neural Network, by L. Xiaoli et al., 8 JOURNAL OF INTELLIGENT MANUFACTURING pgs. 271-276 (1997). In performing a continuous wavelet transformation a portion of the sensor signal is windowed and convolved with a wavelet function. This convolution is performed by superimposing the wavelet function at the beginning of a sample, multiplying the wavelet function with the signal and then integrating the result over the sample period. The result of the integration is scaled and provides the first value for continuous wavelet transform at time equals zero. This point may be then mapped onto a three dimensional plane. The wavelet function is then shifted right and the multiplication and integration steps are repeated to obtain another set of data points which are mapped onto the 3-D space. This process is repeated and the wavelet is moved (convolved) through the entire composite signal. The wavelet is then scaled, which changes the frequency resolution of the transformation, and the above steps are repeated.

Data from a wavelet transformation of a composite sensor signal from sensor 20 is shown in FIG. 4. The data is graphed in three dimensions and forms a surface 41. As shown in the graph of FIG. 4, the composite sensor signal includes a small signal peak at about 1 kHz at time t.sub.1 and another peak at about 100 Hz at time t.sub.2.

In one aspect of the invention, wavelet transformation data such as that represented in FIG. 4 is calculated and stored in memory 28 shown in FIG. 3 during normal operation of the sensor. This data represents a base "plane" of normal operation. The data may be collected at various times during the day, during a process cycle and during the year. When placed into normal use, life expectancy and diagnostic circuitry 34 retrieves the stored wavelet transformation from memory 28 and compares the base plane data with information gathered through wavelet analysis during operation. For example, if circuitry 34 subtracts the base plane data from a current wavelet transformation, the resultant data represents only the anomalies occurring in the process. Such a subtraction process separates the process variations from the sensor signal along with daily and seasonal variations in the signal. For example, the sensor signal may change during the day or over the course of a year due to environmental temperature changes. Thus, this allows separation of the process signal from the signal due to the sensor.

The continuous wavelet transformation described above requires extensive computations. Therefore, in one preferred embodiment, wavelet processing circuit 32 performs a discrete wavelet transform (DWT) which is well suited for implementation in a microprocessor. One efficient discrete wavelet transform uses the Mallat algorithm which is a two channel sub-band coder. The Mallet algorithm provides a series of separated or decomposed signals which are representative of individual frequency components of the original signal. FIG. 5 shows an example of such a system in which an original sensor signal S is decomposed using a sub-band coder of a Mallet algorithm. The signal S has a frequency range from 0 to a maximum of f.sub.MAX. The signal is passed simultaneously through a first high pass filter having a frequency range from 1/2 f.sub.MAX to f.sub.MAX, and a low pass filter having a frequency range from 0 to 1/2 f.sub.MAX. This process is called decomposition. The output from the high pass filter provides "level 1" discrete wavelet transform coefficients. The level 1 coefficients represent the amplitude as a function of time of that portion of the input signal which is between 1/2 f.sub.max and f.sub.MAX. The output from the 0-1/2 f.sub.max low pass filter is passed through subsequent high pass (1/4 f.sub.max -1/2 f.sub.max) and low pass (0-1/4 f.sub.max) filters, as desired, to provide additional levels (beyond "level 1") of discrete wavelet transform coefficients. As shown in FIG. 5, the outputs from each low pass filter may be subjected to further decompositions offering additional levels of discrete wavelet transformation coefficients as desired. This process continues until the desired resolution is achieved or until the number of remaining data samples after a decomposition is too small to yield any further information. The resolution of the wavelet transform may be chosen to be approximately the same as the sensor or the same as the minimum signal resolution required to monitor the process. Thus, each level of DWT coefficients is representative of signal amplitude as a function of time for a given frequency range. Coefficients for each frequency range may be concatenated to form a graph such as that shown in FIG. 4.

FIG. 6 is an example showing a signal S generated by an RTD temperature sensor and the resultant approximation signals yielded in seven levels labelled level 1 through level 7. In this example, signal level 7 is representative of the signal due to the sensor itself and any further decomposition will yield noise. In this particular example, signal due to the sensor is identified as the last signal in the decomposition prior to generating such a noise signal. For example, this can be determined by comparing the differences between successive decompositions and identifying the signal which has the smallest change relative to the next decomposed signal. However, this may vary for different types of sensors or processes.

In some embodiments it may be desirable to add padding to the signal by adding data to the sensor signal near the borders of windows used in the wavelet analysis. This padding reduces distortions in the frequency domain output. This technique may be used with a continuous wavelet transform or a discrete wavelet transform. "Padding" is defined as appending extra data on either side of the current active data window, for example, extra data points are added which extend 25% of the current window beyond either window edge. In one preferred embodiment the extra data is generated by repeating a portion of the data in the current window so that the added data "pads" the existing signal on either side. The entire data set is then fit to a quadratic equation which is used to extrapolate the signal 25% beyond the active data window.

In process control systems where there is a known process variation, for example, due to seasonal changes, the variation can be modeled and thereby removed from the sensor signal to obtain the residual sensor signal. As described above, such modeling may be performed by observing the process or otherwise predicting how the process will change over time. Further, the model may be a function of other process variables or control signals which are used in predicting the process variable signal. In another embodiment of the invention, a number of predetermined models are stored in memory 28. During operation, a neural network operating in microprocessor 26 monitors operation of the process and selects the optimum model stored in memory. Coefficients related to operation of the model may be generated using a neural network or may be received over loop 6 during installation of transmitter 40 as provided for in various communication protocols such as Fieldbus. Examples of models include a first order model including dead time which is typically good for non-oscillatory systems, or second order models plus dead time which typically suffice for oscillatory processes. Another modeling technique is to use an adaptive neural network-fuzzy logic model. Such a hybrid system includes a neural network and fuzzy logic. The fuzzy logic allows adaption of the model to variability of the process while the neural network models allows flexibility of the modeling to thereby adjust to changing processes. This provides a relatively robust model. The use of adaptive membership functions in the fuzzy logic model further allows the determination whether the particular model should be updated.

Further, the novel use of wavelet analysis is well suited for analyzing signals which have transients or other non-stationary characteristics in the time domain. In contrast to Fourier transforms, wavelet analysis retains information in the time domain, i.e., when the signal occurred.

The present invention may operate with any appropriate type of life expectancy or diagnostic circuitry. Examples of such techniques are shown in the co-pending application Ser. No. 08/744,980, filed Nov. 7, 1996, entitled "DIAGNOSTICS FOR RESISTANCE BASED TRANSMITTER," which is incorporated by reference. Further, the invention may be used with any type of process sensor including sensors which measure temperature, pressure, level, flow, pH, turbity, etc.

Although the present invention has been described with reference to preferred embodiments, workers skilled in the art will recognize that changes may be made in form and detail without departing from the spirit and scope of the invention. For example, the sensor may be any type of process variable sensor including temperature, pressure, flow, level, etc.