Application of multi-sensor information fusion based on ARM in industrial control

0 Introduction

Modern industrial production is characterized by comprehensiveness, complexity, large-scale and continuity, and a large number of sensors are used to monitor and control the production process. The emergence of multi-sensor systems has led to a sharp increase in the amount of information, and the use of information fusion technology can make more effective use of information resources. In complex industrial control systems, the control process needs to involve multiple information at the same time, especially the connection between each piece of information, the information characteristics contained in the organic combination of information, and the overall status of the information, and needs to be controlled according to the process operation characteristics described by the comprehensive status. Therefore, the embedded system is combined with information fusion technology to solve problems that are difficult to solve with traditional control.

1 Multi-sensor information fusion

1.1 Concept of Multi-Sensor Information Fusion

Multi-sensor information fusion refers to the coordinated use of multiple sensors to integrate the local incomplete observations provided by multiple sensors distributed in different locations and the relevant information in the associated database, eliminate the redundancy and contradiction that may exist between multi-sensor information, and complement each other to reduce its uncertainty and obtain a consistent description of the object or environment. That is, the data from multiple sensors is processed at multiple levels, aspects and levels to generate new meaningful information, which cannot be obtained by any single sensor.

1.2 Methods of multi-sensor information fusion

Commonly used information fusion methods include weighted average method, Kalman filtering, classical reasoning method, Bayesian estimation, DS evidence decision theory, quality factor method, template method, entropy theory, cluster analysis, fuzzy reasoning, production rules, genetic algorithm, neural network. Among them, the neural network method has a strong information processing ability. For complex industrial intelligent monitoring and control systems and in processing uncertain information, neural networks are a powerful tool, thus providing a good method for information fusion.

The basic idea of ​​using neural networks for information fusion is to determine the classification criteria based on the similarity of samples received by the current system. The determination method is mainly reflected in the distribution of network weights. At the same time, the learning function of the neural network can be used to acquire knowledge and obtain an uncertain reasoning mechanism. The method of the Adaptive Resonance Theory ART (Adaptive Resonance Theory) in the neural network is used. Figure 1 is a network diagram of ART-2 that can process analog information.

Competitive learning mechanism and self-stabilizing learning mechanism are the basis of adaptive resonance theory (ART). The competitive learning mechanism only changes the weight coefficients related to the winner of the competition, while all other weight coefficients remain unchanged. Through learning, the observation vector sets of different objects have found their corresponding winning output components, so they can be naturally classified according to the winner's number.

The self-stabilizing learning mechanism is composed of an information feedback channel, a comparison mechanism and a corresponding algorithm. Its working process is described as follows: 1) competitive selection operation; 2) forming a feedback information channel; 3) comparing similarities; 4) adjusting weight coefficients.

2 Industrial Embedded Control Systems and Information Fusion

2.1 Information Fusion in Industrial Control

In industrial control, the process operation status is referred to as the working condition. In a simple system, the output of a certain sensor can roughly reflect the working condition. In a complex industrial production process, the working condition cannot be directly represented by one or several process variables, and the output of a certain sensor only describes one aspect of the working condition. By adopting appropriate information fusion methods, the information of multiple sensors describing the working condition from different aspects can be fused to obtain a complete description of the working condition, which can be used for operation and real-time intervention, or the system can be automatically controlled according to the working condition. Figure 2 shows the principle of information fusion control, and this process can be decomposed into the following parts:

1) Sensor information acquisition, including multi-sensor system, sensor information preprocessing and soft measurement, and human-machine interface. This part includes hardware and related software, and should directly or indirectly detect information reflecting the operating status and environment of the object as much as possible, including the state quantity of the object, the controlled quantity, and environmental information.

2) Cluster fusion control, which consists of a series of software modules, is the core part of completing intelligent monitoring and control.

3) Interpretation mechanism, including hardware and related software, such as graphics, text, sound, light, multimedia output devices, etc. Explain the current state of the system and the clustering fusion results, and answer questions raised by users through the human-computer interface. In intelligent monitoring systems without automatic control functions, such as fault diagnosis systems and production operation guidance systems, the interpretation mechanism completes all output functions of the system: displaying the current working conditions of the production system, fault diagnosis results, hidden faults and dangerous situation forecasts, production operation suggestions and guidance, etc.

4) The actuator, including related hardware such as power amplifier and actuator, realizes automatic feedback control according to the results of cluster fusion control operation.

2.2 System Hardware Design

The system is mainly composed of the Samsung S3C2410 processor with ARM920T core, external RAM, Flash, D/A conversion chip, LCD and RS232 interface. The S3C2410 chip has an 8-channel 10-bit A/D converter inside, so there is no need to expand the A/D conversion chip. The overall hardware block diagram of the system is shown in Figure 3. [page]

2.3 System Software Design

1) Embedded Linux operating system transplantation

Linux has the feature of open source code. Although the Linux operating system is not a microkernel structure, it uses dynamic module loading, which makes Linux very easy to tailor. The modular structure allows users to easily configure it, remove modules that are not needed by the user system, and reduce system overhead. The embedded Linux system includes a bootloader, a kernel, and a root file system. The steps for developing under the Linux system include: (1) configuring and compiling the bootloader and Linux kernel, and burning them to the target platform; (2) burning the root file system; (3) establishing a cross-compilation environment.

2) Clustering fusion subroutine algorithm

The clustering fusion subroutine is the core of the whole system, which includes the fusion database and expert knowledge base in information fusion, as well as the data and knowledge required in the monitoring, control, calculation and transformation process. Figure 4 is the flow chart of the clustering fusion algorithm.

3) Interface design

Qt/Embedded is produced by TrollTech of Norway. It is a complete C++ application development framework, so Qt is selected for design. Qt contains a class library and tools for cross-platform development. In addition to the C++ library and rich API, QT abstracts external input devices into input events and can support a variety of hardware devices, allowing users to not only develop graphical interface programs, but also control the underlying hardware. Figure 5 is an information fusion control interface that simulates the input of three sensors: boiler water level, flow, and pressure.

3 Conclusion

Multi-sensor information can perfectly, accurately and reliably reflect the characteristics of objects and environments. Introducing information fusion technology into industrial control can make more effective use of information resources. ARM, as an embedded microprocessor with 32-bit RISC architecture, has the characteristics of high integration and high reliability. It can solve the problems of low reliability and poor anti-interference ability in traditional industrial production processes due to the relative dispersion of equipment and devices.