Multi-sensor data fusion fire alarm system based on ARM9

In view of the fact that a single ordinary type of fire detection alarm can no longer meet the demand, the use of multiple sensors to comprehensively collect various abnormal information before a fire occurs, and the use of multi-sensor information fusion technology to process the fire information provided by the sensors can greatly improve the reliability of the entire alarm monitoring system.

1 System Hardware Design

The core controller of the hardware system of this solution is Samsung's 16/32-bit multi-function, low-power embedded processor S3C2440 with ARM920T core. S3C2440 is a high-end embedded microprocessor launched by Samsung in South Korea that can be used for the development of portable products such as industrial control and smart home appliances. Its main frequency processing speed reaches 400MHz, which can fully meet the real-time processing requirements of fire monitoring alarms. Its main control chip and rich peripheral interface circuits can be used to connect various digital devices to realize data exchange. The fire warning system based on multi-sensor data fusion adopts a modular structure, which is mainly composed of sensor module, A/D conversion module, S3C2440 controller, alarm module, execution module and power module and memory module required by the controller. Figure 1 is the system structure diagram.

The multi-sensor module is composed of multiple groups of sensors, each of which is composed of a temperature sensor, a combustible gas detector and a smoke detector. In this system, the temperature sensor uses the HM500 of the French HuMIREL company, which has the characteristics of low cost, small size, long life, good selectivity and stability; the combustible gas detector uses the infrared gas sensor newly developed by Shenzhen Jianda Technology Co., Ltd. The detector is installed in indoor and outdoor dangerous places where the measured gas is easy to leak. They can sensitively sense low-concentration polluted gases in the air, and have high sensitivity to odors, CO, H, and O in the air, and can even detect several ppm-level polluted gas content; the smoke detector uses the American General GE smoke detector 514C, which has self-diagnosis function, drift compensation and dust-induced interference. The above sensors complete the monitoring of multiple parameters of the fire process, transmit the detected data to the S3C2440 controller through the A/D conversion module and are equipped with intelligent discrimination technology, which can achieve the purpose of early warning, reduce missed alarms and false alarms, and improve reliability. The A/D conversion module used in this design is TI's 12-bit high-speed parallel converter ADS805, which has the characteristics of high sampling speed and good stability.

The core of the S3C2440 controller uses a 16/32-bit RISC microprocessor, which adopts a 6-layer board process. It has the characteristics of low power consumption and high-speed processing and computing capabilities. The simple and stable design is very suitable for products with high power requirements. It adopts a new bus architecture (AMBA) and its core is a 32-bit advanced processor. Its main frequency can reach up to 533MHz, which can fully guarantee the real-time requirements when processing a large amount of sensor data. Its power management module can provide the system with multiple voltages, including 1.8V power supply for the chip core voltage and 3.3V power supply for the I/O part of the chip. Some conventional integrated circuits outside the chip are powered by 5V. The intelligent power management module solves the different power supply requirements for various parts of the system, reduces power consumption, reduces interference noise between different power supplies, and improves the integration of the system. Its memory module includes two SDRAMs with a total of 64MB and a 64MNandflash (K9F1208), and other capacity Nandflash memories can be selected according to the storage capacity requirements. The memory module is used to store the system running program and the data collected by the sensor during the fire monitoring process.

The alarm module mainly activates the sound and light alarm signal to notify the on-duty personnel when it determines that there is a fire risk, so that they can take corresponding measures; the execution module activates the nearby fire extinguishing device when the fire risk occurs, so as to minimize the harm of the fire.

2 System Software Design

The software part of the system mainly includes system power-on initialization, system self-test, clock initialization, interrupt setting, peripheral initialization, and then running the main program main() function. After establishing the task, the expansion port controls the ADC to switch channels to collect data. After pre-processing the data such as smoothing, filtering, calibration and compensation, the data fusion algorithm is used to analyze the fire situation, determine whether to alarm and loop detection. The system software flow is shown in Figure 2.

Different from general data acquisition and processing systems. The data acquisition software and data processing software of this system are both run on the processor. In addition to continuously collecting the latest fire scene data, the system software also needs to perform real-time data processing. A measurement and control system with 8/16-bit single-chip microcomputer as the core. The program is generally written in the foreground and background mode. A large infinite loop runs in the background. There are multiple interrupts in the foreground. This method is incapable of ensuring the real-time measurement and control when the program scale increases and the system functions are more complex, especially when there are more concurrent modules in the system. Moreover, programming is difficult and it is not convenient to add functions. Considering the software complexity, computing volume, and real-time requirements, the system adopts the μCOS-II operating system.

Most of the tasks in this system are scheduled to run by calling OSTimeDly(), and each task can be assigned different time intervals through system functions. ADC data acquisition program, data processing program and data fusion algorithm program are mainly written in embedded C language.

3 Data Fusion Algorithm

The application of multi-sensor information fusion technology in industrial process monitoring systems has achieved some engineering applications. In such systems, after the sensors collect data from the object and the environment, they first perform data fusion processing and then participate in the control strategy operation. At present, the commonly used information fusion methods can be roughly divided into the following categories: the first is the classical method based on estimation and statistics, including weighted average method, least squares method and DS evidence theory; the second is the fusion of information theory, including template method, entropy theory of cluster analysis, etc.; the third is the fusion method of artificial intelligence, including fuzzy logic, production rules, neural networks, genetic algorithms, fuzzy integral theory and expert systems.

When applied to multi-sensor information fusion, we regard A as the set of possible decisions of the system and B as the set of sensors. The element μi in the relationship matrix RA+B between A and B represents the possibility of inferring decision i from sensor i, X represents the credibility of the judgment of each sensor, and Y obtained after fuzzy transformation is the possibility of each decision.

Specifically, we assume that there are m sensors observing the system, and the system may have n decisions, then:

A: {y1/decision, y2/decision, …, yn/decision n}
B: {x1/sensor, x2/sensor, …, xn/sensor m}
The judgment of the sensor on each decision is represented by the membership function defined on A. Suppose the judgment result of sensor i on the system is:
μi1/decision, μi2/decision, …, μin/decision n, 0≤μy≤1,
that is, the possibility of the result being decision j is μij, recorded as vector μi1, μi2, μi3, …μin, then the relationship matrix A×B composed of m sensors is:

The credibility of each sensor's judgment is represented by the membership degree on B: X = {x1/sensor 1×x2/sensor 2…, xn/sensor n}, then, according to Y = X * RA * B, fuzzy transformation can be obtained: y = (y1, y2, y3, …, yn)

That is, the probability of each decision after comprehensive judgment is y. Finally, each possible decision is selected according to certain criteria (such as the maximum membership method, the center method, etc.) to obtain the best result. According to the calculated y value, the following rules should be used for judgment: ① The judgment result should have the maximum membership. ② The membership of the judgment result must be greater than a certain threshold (generally 0.5). ③ The difference between the membership of the judgment result and the membership of other judgments must be greater than a certain threshold (such as 0.1).

4 Data fusion experiment of fire monitoring

This design uses temperature sensors, combustible gas detectors and smoke detectors for fire monitoring. The data fusion method is shown in Figure 3.

Figure 3 The general method of data fusion based on fuzzy reasoning is that in the fire fault monitoring system, the weight of each sensor is first determined. In the design, we set the weights of the temperature sensor and the combustible gas detector to W1=0.5, W2=0.3, and W3=0.2 respectively; the final judgment results are divided into two types: fire Y1 and no fire Y2; according to the current working status, the membership function of each sensor X for each judgment Y is determined; and then a linear transformation operation is performed to determine the final result.

For example, at a certain moment, the membership of whether there is a fire or not according to the data of the temperature sensor is μ11=0.45, μ12=0.55, and the membership of whether there is a fire or not according to the data of the smoke sensor is μ31=0.9, μ32=0.1. The linear transformation operation is used to obtain Y,

According to the results, there is a fire hazard and the aerosol fire extinguisher should be activated. Table 1 is the test data of fuzzy fusion in the fire fault monitoring system.

5 Conclusion

The fuzzy reasoning data fusion method is applied to the multi-sensor cable fire fault monitoring system, which has higher accuracy and credibility compared with a single sensor. The operation results show that this method is practical and effective in improving the reliability of fire fault detection and can reduce the false alarm rate of fire alarms. However, this method also has some disadvantages, such as the weight of the sensor and the allocation of the membership value of each sensor to the judgment. There is no unified theory, which needs to be set based on experience.