Application of Wireless Sensor Network in Monitoring of Weapon and Equipment Production

Weapon production [1-2] requires intricate process equipment that contains various large and medium compressors, pumps, pressure vessels, motors, etc. The production environment is featured by high temperature, pressure, flammability, explosibility and toxicity. The process equipment must be monitored effectively to prevent accidents, reduce downtime rate, extend service life and control maintenance cost.

The operating condition of weapon production equipment is traditionally monitored by wired sensor control system. However, this traditional system cannot cover a wide area without the deployment of numerous sensors in different types. This defect greatly constrains the application of wired sensor control system in the process industry.

The wired sensor control system is gradually being replaced with wireless sensor network (WSN) [3-4], an emerging tool for information acquisition and processing. Focusing on data processing, the WSN can effectively overcome the defect of wired sensor control system and achieve real-time monitoring of the operating condition of weapon production equipment.

The monitoring of the operating condition needs to capture various operating parameters of the equipment, including but not limited to temperature, pressure and vibration. The analysis on these parameters helps to acquire, analyze and evaluate the status of the equipment, laying the basis for correct judgement of abnormalities. Suffice it to say that operating condition monitoring is the prerequisite for equipment monitoring and fault diagnosis.

There are three ways to monitor the operating condition, namely, offline periodic detection, online monitoring and offline analysis, and automatic online monitoring and analysis [5]. Over the years, the equipment monitoring technology has been improved constantly towards automation, networking and virtualization. The scale and applicable range of the monitoring systems are expanding continuously, forming intelligent computer network monitoring systems with such abilities as rapid data acquisition, real-time online analysis, and distributed hierarchical distributed structure.

To solve the defect of traditional wired sensor control system, this paper designs a data acquisition and monitoring system for the operating condition of weapon production equipment based on the WSN, and proves through experiment that the proposed system outperforms the traditional system in acquiring the data and monitoring the operation of weapon production equipment.

Real-time monitoring is essential to the stable and safe operation of weapon production equipment. To meet the real-time requirement, it is necessary to continuously capture and transit the feature signals of the equipment operating condition. Due to the diversity of equipment, the data acquisition will inevitably produce a large number of sensor data. This calls for the fusion of multi-sensor data without reducing the real-time performance, such as to reduce the computing and transmission load and improve the network utilization. The purpose of data fusion is to eliminate the measurement uncertainty according to a limited amount of collected data. The fused operating parameters can reflect the equipment condition more accurately.

3.1 Batch estimation fusion algorithm based on weighted mean

In general, the sensor parameters change slowly and obey normal distribution, while the measured data are theoretically unequal in precision. Thus, the traditional normalization method of taking the arithmetic mean is insufficient to ensure the data accuracy. Here, the batch estimation based on weighted mean is introduced to normalize the measured data [7].

To prevent node failure and enhance data reliability, two or more homogeneous sensor nodes were considered to monitor an equipment. After all, a single sensor node has a limited energy. Let n be the number of consistent data collected by the multiple sensors in a period of time. Then, the measured data can be expressed as x₁, x₂,…,x_n. The n measured data were divided randomly into two groups, and the mean value of each group was computed. Next, a batch estimation algorithm was introduced to determine the fusion value X_p according to the computed mean value, making it close to the measured true value. The process can also be considered as two sets of consistency measured data collected by two homogeneous sensors in the same time period.

Assuming that m+k=n, the first and second sets of consistency data series can be denoted as X₂₁, X₂₂,…, X_2kand X₁₁, X₁₂,…, X_1m, respectively. Then, the arithmetic mean and standard variance of the two groups of measured data can be respectively defined as:

$\overline{X}_{(1)}=\frac{1}{m} \sum_{i=1}^{m} X_{1 i}$;

$\sigma_{(1)}=\sqrt{\frac{1}{m-1} \sum_{i=1}^{m}\left(X_{1 i}-\overline{X}_{(1)}\right)^{2}}$;

$\overline{X}_{(2)}=\frac{1}{k} \sum_{j=1}^{k} X_{2 j}$;

$\sigma_{(2)}=\sqrt{\frac{1}{k-1} \sum_{j=1}^{k}\left(X_{2 j}-\overline{X}_{(2)}\right)^{2}}$.

The variance estimates of means $\overline{X}_{(1)}$ and $\overline{X}_{(2)}$ are:

$\sigma_{\overline{X}_{(1)}}^{2}=\frac{1}{m} \sigma_{(1)}^{2}$;

$\sigma_{\overline{X}_{(2)}}^{2}=\frac{1}{k} \sigma_{(2)}^{2}$.

The weight ratios of the two sets of data are:

$\frac{p_{(1)}}{p_{(2)}}=\frac{\sigma_{\overline{X}_{(1)}}^{2}}{\sigma_{\overline{X}_{(2)}}^{2}}$.

Thus,

$\overline{X}_{p}=\frac{\sigma_{\overline{X}(2)}^{2}}{\sigma_{X_{(1)}}^{2}+\sigma_{\overline{X}_{(2)}}^{2}} \overline{X}_{(1)}+\frac{\sigma_{\overline{X}_{(1)}}^{2}}{\sigma_{\overline{X}_{(1)}}^{2}+\sigma_{\overline{X}_{(2)}}^{2}} \overline{X}_{(2)}$.

The corresponding variance estimates are:

$\sigma_{\overline{X}_{p}}^{2}=\frac{\sum_{i=1}^{2} p_{(i)}\left(\overline{X}_{(i)}-\overline{X}_{p}\right)^{2}}{\sum_{i=1}^{2} p_{(i)}}$.

The operating parameters and measured noise of the sensors in our system can be considered as normally distributed, which are suitable for the batch estimation fusion algorithm based on weighted mean.

3.2 Data fusion algorithm based on Haar wavelet transform

The Haar wavelet transform aims to reduce the communication data of sensor nodes. The data fusion algorithm can be established based on Harr wavelet transform in the following manner.

To begin with, the sensor nodes sampling is regarded as a data flow pipeline, the length of which is 2^j units. One of these units is the space needed to store the data structure of a real-time sampled data. It is assumed that the length of the pipe is 2³, and can be treated as a vector space of V³. Then, V³=V²⊕W². Thus, the data volume can be reduced by continuously reducing the dimensions of vector space.

The data of a sensor node can be fused with the vector space of V³through the following steps:

Step 1: Through linear combination of Haar functions, the vector space V³ is represented as $I(x)=α_0^3 φ_0^3 (x)+α_1^3 φ_1^3 (x)+α_2^3 φ_2^3 (x)+…+α_7^3 φ_7^3 (x)$, where $α_0^3$, $α_1^3$, …, $α_7^3$ are sensor sampling values at the fusion degree of 8.

Step 2: The dimension of V³ is reduced to V². Here, V³ is expressed as a linear combination of the basis functions of V² and W²:

$I(x)=α_0^2 φ_0^2 (x)+α_1^2 φ_1^2 (x)+α_2^2 φ_2^2 (x)+α_3^2 φ_3^2 (x)+β_0^2 ϑ_0^2 (x)+β_1^2 ϑ_1^2 (x)+β_2^2 ϑ_2^2 (x)+β_3^2 ϑ_3^2 (x)$;

V²: $I(x)=α_0^2 φ_0^2 (x)+α_1^2 φ_1^2 (x)+α_2^2 φ_2^2 (x)+α_3^2 φ_3^2 (x)$;

W²: $I(x)=β_0^2 ϑ_0^2 (x)+β_1^2 ϑ_1^2 (x)+β_2^2 ϑ_2^2 (x)+β_3^2 ϑ_3^2 (x)$;

where $α_0^2$, $α_1^2$, $α_2^2$ and $α_3^2$ are sensor sampling values at the fusion degree of 4; $β_0^2$, $β_1^2$, $β_2^2$ and $β_3^2$ are detail coefficients.

By this fusion algorithm, the vector space of V³ can be described by function $φ_0^0$, $ϑ_0^0$, $ϑ_0^1$, $ϑ_1^1$, $ϑ_0^2$, $ϑ_1^2$ and $ϑ_3^2$.