SIGNAL PRE-PROCESSING SUBSYSTEM FOR THE PURPOSE

OF INDUSTRIAL CONTROL

Ivan Puchr

COMPUREG Plzeň, s.r.o., Nádražní 18, Plzeň, Czech Republic

Pavel Herout

Department of Computer Science and Engineering, University of West Bohemia, Plzeň, Czech Republic

Keywords: Signal processing, Real-time systems, Control systems, Rolling mill.

Abstract: Commonly used methods of signal processing are often too complex and many of them induce time delay

between the original and processed signal. A problem arises how to supply the controller with quality input

signals without substantial delay taking limited computing power of the controller's hardware into account.

A possible solution based on a signal pre-processing subsystem (SPS) delivering quality input signals to the

controller is described in this paper. Concept of the SPS is based on a hardware unit with a real-time

operating system managing a composition of application tasks. Hardware and software structure of the SPS

is composed to enable to utilize the following specific signal processing strategy. Different signal

processing methods process a particular signal simultaneously and an algorithm of continuous decision

making provides the momentary best possible result in the form of a weighted combination of outputs

generated by single signal processing methods. Solution is intended for enhancement of the automatic

thickness/gauge control (AGC) for cold rolling mills.

1 INTRODUCTION

The quality of a process control is substantially

influenced by the quality of the controller's input

signals. In cases where the quality of input signals is

low, methods of signal processing come into

account. But these methods are often complex and

require a computing power. At the same time the

methods of signal processing (e.g. filtering) usually

bring some time delay of the signal (Vaseghi, 1987).

If the controlled process is fast, quality of input

signals is low and the control algorithm is complex

at the same time, a problem arises how to deliver

quality input signals without substantial delay to the

control algorithm with a limited computing power of

the controller's hardware. With the aim to solve this

problem, the computing power is usually increased

within the controller's hardware unit by the use of

higher clock frequency of CPU or by adding of

another CPU or another CPU board. But there are

cases where this solution is not possible without the

change of the whole hardware base or where current

technical limits are reached. Then, another

possibility is used. The controller's hardware is

distributed to more hardware units connected to a

computer network.

This paper describes the latter possibility - a

solution of this problem using a signal pre-

processing subsystem (SPS). The SPS is designed

for delivering quality input signals for the controller

via a computer network. Solution is introduced in

the application of the automatic thickness/gauge

control (AGC) of a metal strip produced by a cold

rolling mill.

2 MOTIVATION

Some industrial applications of control indicate that

the standard concept using an all-in-one hardware

unit for input signal acquisition, control algorithm

execution and output signal distribution has in some

cases its drawbacks:

 While the control algorithm and the part of the

controller ensuring the algorithm execution are

more or less standardized for a particular

415

Puchr I. and Herout P..

SIGNAL PRE-PROCESSING SUBSYSTEM FOR THE PURPOSE OF INDUSTRIAL CONTROL.

DOI: 10.5220/0003530404150418

In Proceedings of the 8th International Conference on Informatics in Control, Automation and Robotics (ICINCO-2011), pages 415-418

ISBN: 978-989-8425-74-4

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

application, the part of the controller supplying

the control algorithm with input signals is highly

dependent on the nature and quality of input

signals and usually has to be changed

substantially.

 Computing power of controller's hardware unit

may not be sufficient for execution of both the

complex control algorithm (Ettler, 1992) and

application of sophisticated signal processing

methods on low-quality input signals.

 Because of an effort to minimize time delay

imposed by some signal processing methods on

the signals, the scan frequencies of signal

acquisition must be much higher than the actual

frequency of control loop, and this brings further

demands on computing power of the hardware

unit.

 Last but not least, the all-in-one hardware unit

solution does not allow to position the input

signal data acquisition and digitization as near as

possible to the sensor which is desirable to avoid

long distance wiring and to reduce noise

influence.

3 SOLUTION

Possible solution of the above mentioned drawbacks

includes decomposition and functional and

topological distribution of the controller's hardware

unit. This approach is widely spread and is

commonly used in many control applications. Main

contribution of our solution consists in complex

approach to the processing of input signals with the

aim to find a base for a more or less universal

solution of input signal quality improvement in a set

of similar control applications.

3.1 Decomposition and Distribution

The original all-in-one solution of the controller is to

be decomposed first. The input signal acquisition is

detached from the controller's hardware unit and

realized by a separate hardware unit called signal

pre-processing subsystem (SPS). Generally, there

can be a set of SPSs in the framework of a particular

control application. The control algorithm and

output signal distribution is executed within the

original hardware unit.

The SPS is located as near as possible to the

sources of the acquired input signals.

The SPS processes the inputs and resulting

enhanced signals are transferred to the controller via

a dedicated high speed computer network.

3.2 Signal Pre-processing Subsystem

The signal pre-processing subsystem (SPS) is the

new key component of the controller.

3.2.1 Concept of Signal Processing in SPS

Processing of controller's input signals in the SPS

has the following key ideas:

 Scan frequencies of input signal acquisition are

an order higher than the frequency of control

algorithm loop.

 Acquired signal data is buffered before further

processing.

 One input signal is processed by several signal

processing methods in parallel.

 The currently best result of the particular signal

processing methods or a weighted combination

of several ones is chosen as final input to the

controller with the help of a probabilistic

algorithm of continuous decision making.

 The latest values of processed signals are

transmitted via the network to the controller

upon request.

Real-time operating system is the base for the

application software structure. Structure and code of

the application tasks are kept as simple as possible to

spare the computing power and enable to increase

the scan frequencies of the SPS.

The increase of the signal acquisition scan

frequencies brings obviously better starting

conditions for subsequent signal processing

methods. Contribution of the increased scan

frequency of the SPS in comparison to the input

signal acquisition frequency of the original all-in-

one controller is illustrated by Figure 3.

Sufficient computing power of the SPS dedicated

to signal processing enables to process one signal by

several signal processing methods in parallel.

The selection and combination of the outputs

from the particular signal processing methods is the

matter of a specialized probabilistic algorithm of

continuous decision making. Outputs from this

algorithm are buffered in the SPS and prepared for

transmission to the controller.

3.2.2 Transmission of Pre-processed Input

Signals from the SPS to the Controller

As mentioned above, the pre-processed input signals

are ready and continuously updated in the SPS's

memory. For the transmission of the processed

signals to the controller, a local area network (LAN)

is used. It can bring difficulties together with

ICINCO 2011 - 8th International Conference on Informatics in Control, Automation and Robotics

416

Figure 1: Internal structure of the signal pre-processing

subsystem.

advantages mentioned above. Possible unreliability

and time delay of data transmission via the LAN

must be taken into account. These problems are

discussed in many papers concerning the networked

control systems. A method for testing of networked

control systems is described in (Hassapis et al.,

2005).

For the purposes of testing and proof of

functionality of the SPS concept, a common type of

network is used. SPSs and the controller are

connected by standard Gigabit Ethernet network.

This type of network can ensure sufficient stability

of the scan period under following conditions:

 The amount of transmitted data is relatively

small.

 The communication has the character of

frequently transmitted short packets.

 The network is dedicated to SPSs and controller

only.

 Controller acts as a logical master and triggers

communication with SPSs - slaves.

 A low level communication protocol is used (e.g.

TCP/IP sockets).

An alternative to this simple solution is one of

the real-time industrial networks.

4 PILOT APPLICATION

As the pilot application of the SPS, the automatic

thickness/gauge control (AGC) of a metal strip

produced by a cold rolling mill was selected.

Figure 2: An example of a cold rolling mill.

This is an application where principles of the

technological process are well known and control

algorithms are proved. Further improvements of

control algorithms bring only small increase of the

production quality. On the other hand, there is a big

potential in the quality improvement of input signals

that positively influences control algorithm results

and thus the production quality can be increased.

The following picture illustrates the contribution

of higher signal acquisition frequency of the SPS.

The signal is the strip speed measured by an

incremental rotary encoder.

32 600 32 650 32 700 32 750 32 800 32 850 32 900 32 950 33 000 33 050 33 100

0.902

0.904

0.906

0.908

0.91

Speed [m/s]

Sample number

Fast (thin line) and standard (thick line) signal acquisition frequency

Figure 3: Comparison of raw signal data acquired by the

original all-in-one controller (thick line) and by the SPS

(thin line). Sample number of X-axis corresponds to the

thin line.

In the following sequence of pictures, there can

be seen an angular speed signal processed by the

SPS. The signal measures the rotation of the

deflection roll using an incremental rotary encoder.

The signal is processed by three methods that are

executed simultaneously. The first method is a

moving average filter with the window

corresponding to one revolution of the roll. The

second method is a trend preserving filter designed

for a short delay filtering. The third method

SPS

Input boards

Signal m

acquisition

Signal 1

processing

method 1

Raw signal buffers

. . .

...

Communication controller

Processed signal

buffers

Algorithm of

continuous decision

...

Signal 1 Signal m

Signal 1

acquisition

Signal 1

processing

method n

Signal m

processing

method 1

Signal m

processing

method n

Algorithm of

continuous decision

SIGNAL PRE-PROCESSING SUBSYSTEM FOR THE PURPOSE OF INDUSTRIAL CONTROL

417

calculates a correction of filtered signal that

improves the result especially during the time of

acceleration and deceleration. This method is based

on the incremental rotary encoder's properties. For

detailed description see (Ettler et al., 2010). These

methods are processed in parallel, and in this

selected time period, a combination of the second

and the third method is selected as the result.

2000 3000 4000 5000 6000 7000 8000 9000 10000

3.5

3.55

3.6

[rad/s]

raw signal

Sample number

Figure 4: Angular speed raw signal.

2000 3000 4000 5000 6000 7000 8000 9000 10000

3.5

3.55

3.6

Sample number

[rad/s]

filtered by moving average filter (method1)

Figure 5: Angular speed filtered by moving average filter.

2000 3000 4000 5000 6000 7000 8000 9000 10000

3.5

3.55

3.6

Sample number

[rad/s]

filtered by trend preserving filter (method2)

Figure 6: Angular speed filtered by trend preserving filter.

2000 3000 4000 5000 6000 7000 8000 9000 10000

-0.02

0.02

Sample number

corr

[rad/s]

acceleration / deceleration correction (method3)

Figure 7: Correction of the filtered angular speed that

improves the result especially during the time of

acceleration and deceleration.

2000 3000 4000 5000 6000 7000 8000 9000 10000

3.5

3.55

3.6

Sample number

TPcorr

[rad/s]

Combination of method2 and method3 (

corr

)

Figure 8: Combination of two methods selected as the

input for the controller.

1300 1400 1500 1600 1700 1800 1900

3.55

3.555

3.56

3.565

(- -),

TPcorr

[rad/s]

Combination of

corr

in comparison to

- detail

Sample number

Figure 9: Comparison of the trend preserving filter output

with and without (dotted line) acceleration / deceleration

correction.

5 CONCLUSIONS

With the aim to improve the results of a set of high

speed control applications, the signal pre-processing

subsystem (SPS) was designed. The functionality of

the controller with SPS was approved by the pilot

application of the automatic thickness/gauge control

(AGC) of a metal strip produced by a cold rolling

mill. Several basic signal processing methods

specific for the selected application were

implemented within the SPS. Development and

implementation of further methods and of the

selection and combination algorithm are a part of

two currently running research projects.

REFERENCES

Vaseghi, S.V., 1987. Advanced Digital Signal Processing

and Noise Reduction, John Wiley & Sons. New Jersey.

Ettler, P., 1992. Advanced algorithms contra lack of

computing power. In Proceedings of IFAC Workshop

on Mutual Impact of Computing Power and Control

Theory, Prague, Czech Republic.

Ettler, P., Puchr, I., Štika, J., 2010. Combined Approach

Helping to Reduce Periodic Disturbances in Speed

Measuring. In PSYCO 2010, IFAC International

Workshop on Periodic Control Systems, Antalya, TR.

Hassapis, G., Geronatsios, S., Grigoriadis, J. , 2005.

Testbed Evaluation of Networked Control Systems. In

proceedings of ICINCO 2005, 2

International

Conference on Informatics in Control, Automation and

Robotics, Barcelona, Spain.

ICINCO 2011 - 8th International Conference on Informatics in Control, Automation and Robotics

418