Research On System Of Gas Sensor Used In Mine Based On TDLAS

For a long time, the proportion of gas accident which occupies in the total number of coal mine accident in our country stays continuously at a high level because the natural condition is bad and the technology of gas detection is backward. And the number of casualties which gas accident brings about accounts for more than 90% of that of extraordinarily big accident. Therefore, developing a new generation of gas sensor has the vital significance in the production safety of the mine.At present, gas sensor used in the mine adopts almost the catalytic combustion technology, which has many shortcomings, such as, short existence service's life, bad working stability, adjustment frequently, narrow detection range. With the successful developing of distributed feedback laser (DFB laser), the technology of gas detection based on tunable diode laser absorption spectroscopy (TDLAS) has attracted increasing attention. TDLAS is the technology of trace species detection, which detects the vibrational-rotational absorption spectrum within special molecular spectrum. It has many characteristics, such as high sensitivity, high precision, strong selective, quick response speed, and its detection limit may achieve the order of 10-6 ~10-7. Existing gas sensor has many shortcomings that it only can realize detection at a single node, its detection range is narrow, only can provide real-time data, can not carry out the forecast on the future gas concentration. To solve above problems, this paper has studied on the technology of gas detection based on TDLAS with the open long path. The main research content and the achievement are showed as follows:1. Harmonic in the Lorentz linetype with the Fourier's series has been analyzed. The result indicated that, in the peak position of the gas spectral absorption, the amplitude of the odd harmonic is zero, and that of the even harmonic has the maximum value. The amplitude of the low harmonic is bigger than that of the high harmonic as a whole. With the increasing of amplitude, the spectral line width will become broader gradually, and its amplitude presents the tendency that increases firstly and then reduces. The maximum amplitude of the even harmonic from 2 to 12 respectively is 0.343, 0.180, 0.122, 0.092, 0.073, 0.060, which corresponding best modulation amplitude respectively is 2.197, 4.116, 6.081, 8.061, 10.049, 12.041.2. Influence on harmonic under the residual amplitude modulation (RAM) and interference with Bessel functions has been analyzed. The results showed that, background signals caused by RAM and interference is proportional to the laser's output light intensity and its modulation factor. RAM effects majorly first and second harmonic, and it may restrain the influence on harmonic caused by RAM through adjusting phase difference between the reference signal and measured signal in LIA. Background signals caused by interference depends on Bessel functions. By selecting an appropriate amplitude of frequency modulationδωthat makes J n(δωτ)=0, it can restrain the influence caused by interference.3. The system structure and unit electric circuit has been designed. Remote gas sensor structure of optical path has been improved by angular reflector, to detect the mean gas concentration in the certain region has been realized. It realized the scanning of the spectral absorption lines using the method of the modulation signal superimposing on the low frequency triangular wave. The electric circuit has been designed, such as pre-amplifier. CAN Bus node and converter between RS232 and CAN has been developed. By establishing the communication protocol of the CAN user, the interconnection among gas detection nodes in the mine has been realized.4. The technology of the harmonic's extraction and denoise has been researched. Digital LIA (DLIA) has been developed and influence on frequency drift in DLIA has been analyzed. The electric circuit, such as modulating signal generator, programmable amplifier, filter, data acquisition system, has been designed. Reference signal produced by DDS and measured signal are cross correlated, which can extract the signal submerged in noise. The result indicated that, DLIA can restrain most of noise in measured signal. The output from DLIA responds to harmonic of the reference signal with same frequency, while it doesn't almost respond to that with different frequency. Owing to influence caused by RAM and DLIA itself, the output from DLIA still included noise in certain degree. So, signal must be denoised with the technology of adaptive wavelet threshold filtering. As a result, after wavelet filtering, the pure first and second harmonic can be got, and the signal-to-noise (SNR) of harmonic is higher, the filter effect is better.5. The system's detection precision, linearity and range with piece linear interpolation has been improved. The former analysis and design all is under the condition ofα( v)CL<<1, and it can approximate Lambert-Bear's expression with linear function. This method has many problems mainly as follows:If gas absorption path is long, when gas concentration is low, it can strengthen the light absorption, then precision and linearity is high; when gas concentration is high, the value ofα(v)CL is big, and the inequality ofα( v)CL<<1 is untenable. It can cause big detection error, even detection wrong, so as to restrain the detection range in the narrow range. If gas absorption path is short, when gas concentration is high, it can also satisfyα( v)CL<<1, then precision and linearity is high, the range is wide; when gas concentration is low, the light absorption is very weak, the SNR of measured signal is very low, which can cause bad influence on the system's precision.When it carried out the analysis adopted the wavelength at 1650nm and 1331nm, it can get the result that if the absorption path is longer, concentration is higher, and absorption coefficient is bigger, when Lambert-Bear's expression approximated by linear function, then error produced by this method is bigger, the nonlinearity is bigger, and the range is narrower.In view of above situation, by approximating Lambert-Bear's expression with the function of piecewise linear interpolation and the introduction of proportionality coefficient, it can solve the problems caused by the untenable situation of the inequality ofα( v)CL<<1. When selected weak 1331nm absorption line of gas to detect, the result showed that, on the condition that the length of gas absorption path is 100cm, when gas concentration is low, detection precision and linearity is high no matter adopting piecewise linear interpolation or linearity function. With the increasing of gas concentration, the error caused by using the linear function became big. When gas concentration is bigger than 10%, it will appear the error value. Detection range can reach 0~100%, the relative error in the entire detection range is lower than 3%, the linearly dependent coefficient between measured value and true value can achieve 0.99958.6. Gas concentration forecast based on BP neural network has been realized. In many situations, the gas explosive is just because gas surged out suddenly. However, the existing sensor can almost only provide the real-time gas concentration data, while they cannot reflect accurately the trend of the change of gas concentration. To solve this problem, by establishing the prediction model of gas concentration based on BP neural network, and the performance of BP neural network adopted adaptive to make sure the nodes of hidden layer can be improved. Adopted training function, such as algorithm with batch gradient descent, algorithm with self-adaption modified learning rate, algorithm with Levenberg-Marquardt, BP algorithm with rebound, algorithm with scaled conjugate gradient, and algorithm with Quasi-Newton, the prediction model can be trained using 30 groups samples. The results indicated that, the training speed adopted by algorithm with Levenberg-Marquardt is quickest. The forecast error of gas concentration by model is smaller than 6%. When the change of input data is not big, the forecast error of model is small, while it is very big when the change is big. The model forecasts the future gas concentration just through the known gas concentration in the period of time, so the dependence on the historical data is very big.Main innovative work is showed as follows:1. To solve the problem of the untenable situation of the inequality ofα( v)CL<<1, the method of piecewise linear interpolation has been put up with. By this method, the system's detection precision, linearity and detection range can be improved.2. To detect the mean gas concentration in a certain area by improving open long path with corner reflector has been realized3. The forecast model of gas concentration has been established based on BP neutral network. The future gas concentration can be forecasted with single historical data. The complexity of the model has been reduced, and speed and stability of forecast has been improved.The existing problem is mainly showed as follows:1. Precision by the forecast model of gas concentration is not high.2. Spot test doesn't be carried on.3. There is no treatment of anti-explosion.