An Adaptive System for Multiple Gas Sensors and Its Parameter Calculation Method

Abstract

The present invention discloses an adaptive system for multiple gas sensors and a parameter calculation method thereof, belonging to the technical field of gas detection. The system includes a front-end polarity adaptation unit for connecting to a gas sensor and receiving the current signal of the gas sensor, a current-voltage conversion unit, an intelligent neural network calculation unit U5, and an analog-digital conversion unit U4. The signal after front-end polarity matching and current-voltage conversion is subjected to signal filtering and signal gain amplification by the intelligent neural network calculation unit U5, and the analog-digital conversion unit U4 performs the conversion between analog and digital quantities to obtain the detected gas voltage value of the gas sensor. The intelligent neural network calculation unit U5 calculates the current gas concentration. The present invention is adaptable to gas sensors of different manufacturers, different principles, and different ranges, and realizes intelligent adaptation for multi-gas detection without increasing the product cost; it can calculate accurate concentrations for different gases.

Description

Technical Field

[0001] The present invention belongs to the technical field of gas detection, and in particular relates to an adaptive system for multiple gas sensors. Background Art

[0002] Gas sensor types include photoionization gas sensors, semiconductor gas sensors, contact combustion gas sensors, electrochemical gas sensors, non-dispersive infrared gas sensors and many other types of sensors. The multi-gas detectors currently on the market often adopt a fixed detection object and a one-time configuration. Different types of gas sensors cannot be replaced later, and the detection objects cannot be expanded after leaving the factory. In order to be compatible with more detection objects, the equipment needs to add a lot of sensor hardware and software resources. Some functions are not used on site, resulting in waste of resources and increased costs. These increased costs are ultimately passed on to end customers, causing price increases.

[0003] Existing technologies often achieve multi-gas detection by adding more sensors to the device. Limited by size and cost, these devices often employ semiconductor sensors or MEMS sensor arrays. However, cross-interference between gases within these sensors can lead to low accuracy in actual detection data, severely impacting practical performance. Another approach involves replacing sensors with the appropriate ones for different on-site detection targets.

[0004] There are two implementation routes in the existing technology:

[0005] Technical Route 1: The Chinese utility model patent with publication number CN219391885U discloses a multi-scenario environment adaptive gas semiconductor array gas sensing device that uses a semiconductor sensor array to deploy multiple sensors at one time to detect multiple gases.

[0006] The problems with this existing technology are: placing multiple sensors at a time increases the cost; using semiconductor sensors has poor accuracy and large temperature drift, making it impossible to accurately detect ambient gases. It can only be used as an alarm and has limited application scenarios.

[0007] Technical route 2:

[0008] The Chinese invention patent application with the publication number CN118090884A discloses a weak signal measurement system for an adaptive gas sensor, which uses a first-stage transimpedance amplifier circuit, a second-stage amplifier circuit, and an analog filter circuit, and uses an MCU unit to control an analog switch to achieve gain control of the second-stage amplifier circuit; the MCU unit reads the digital signal converted by the differential AD converter and compares it with the internal threshold value, and controls the analog switch to switch and control the resistance of the resistance amplification network connected to the second-stage amplifier circuit to adjust the amplification factor of the second-stage amplifier circuit. The technical solution of this patent application can only be adapted to different ranges of the same type of sensor, and can meet the amplification of small signals through gain adjustment; however, in actual use, due to different principles of gas sensors, the polarities of the output signals of gas sensors are positive and negative, and the corresponding amplifier circuits are also different; for gas sensors with different polarities, the technical solution of this application cannot adaptively match the polarities of the amplifier circuits. Summary of the Invention

[0009] In view of the above problems existing in the prior art, the technical problem to be solved by the present invention is to provide an adaptive system for multiple gas sensors and its parameter calculation method, which can adapt to different types of sensors. For different gases to be detected, only the corresponding gas sensor needs to be replaced, which is convenient for realizing low-cost intelligent online monitoring of multiple gases.

[0010] Technical Solution: To solve the above technical problems, the technical solution adopted by the present invention is as follows:

[0011] An adaptive system for multiple gas sensors includes a front-end polarity adaptation unit for connecting to a gas sensor and receiving the current signal of the gas sensor, a current-voltage conversion unit connected to the front-end polarity adaptation unit, an intelligent neural network calculation unit U5 connected to the current-voltage conversion unit, and an analog-to-digital conversion unit U4 connected to the intelligent neural network calculation unit U5. The intelligent neural network calculation unit U5 realizes signal front-end polarity matching by controlling the front-end polarity adaptation unit. The intelligent neural network calculation unit U5 converts the current signal of the gas sensor into a voltage signal by controlling the current-voltage conversion unit. The signal after front-end polarity matching and current-voltage conversion is filtered and amplified in signal gain by the intelligent neural network calculation unit U5, and then sent to the analog-to-digital conversion unit U4. The intelligent neural network calculation unit U5 controls the analog-to-digital conversion unit U4 to perform the conversion between analog and digital quantities to obtain the detected gas voltage value of the gas sensor, and the intelligent neural network calculation unit U5 calculates the current gas concentration.

[0012] Preferably, the front-end polarity adaptation unit includes a first polarity transformation unit U2-1 and a second polarity transformation unit U2-2. The current-voltage conversion unit includes a transconductance amplifier U1A, a preamplifier U1B connected to the transconductance amplifier U1A, and a digital resistor U3 connected to the preamplifier U1B. The first polarity transformation unit U2-1 is connected to the transconductance amplifier U1A through a first resistor R1B, and the second polarity transformation unit U2-2 is connected to the transconductance amplifier U1A through a second resistor R2B. The intelligent neural network calculation unit U5 adapts to the type of gas sensor, activates and switches the first polarity transformation unit U2-1 and the second polarity transformation unit U2-2, and controls the analog switch polarity matching of the front-end polarity adaptation unit. The gas sensor current signal is connected to the transconductance amplifier U1A through the first polarity transformation unit U2-1 or the second polarity transformation unit U2-2. The transconductance amplifier U1A is connected to the preamplifier U1B through a third resistor R3. The gas sensor current signal is converted into a voltage signal through the third resistor R3 and is primarily amplified by the preamplifier U1B. The inverting input terminal of the preamplifier U1B is connected to a fourth resistor R4 and is connected to the digital resistor U3 through a sixth resistor R6. The intelligent neural network calculation unit U5 automatically calculates the gain of the primary amplification based on the type, range, and manufacturer information of the gas sensor, and controls the output resistance of the digital resistor U3. The output terminal of the preamplifier U1B is connected to the intelligent neural network calculation unit U5. The voltage signal output by the preamplifier U1B undergoes digital filtering and secondary amplification by the intelligent neural network calculation unit U5 to meet the input requirements of the analog-to-digital conversion unit U4. The intelligent neural network calculation unit U5 controls the analog-to-digital conversion unit U4 to perform the conversion between analog and digital quantities, obtaining the detected gas voltage value of the gas sensor. The intelligent neural network calculation unit U5 automatically calculates the sensitivity parameter, sensor environment calibration parameter, and zero point parameter of the gas sensor based on the gas sensor type, range, and manufacturer information, and finally calculates the current gas concentration.

[0013] Preferably, it further includes a display unit U6, an alarm unit U7, and a remote terminal unit U8 connected to the intelligent neural network calculation unit U5. The display unit U6 is used to display the gas detection concentration display and the alarm display. The alarm unit U7 is used for over-limit early warning. The remote terminal unit U8 is used for remote data interaction, obtaining the gas sensor type, range, and manufacturer information, and synchronizing the detected gas concentration and alarm information to the remote data center.

[0014] Preferably, the intelligent neural network calculation unit U5 includes a first part unit and an intelligent parameter operation unit. The first part unit is used for signal filtering and amplification, and the intelligent parameter operation unit is used for parameter calculation of the gas sensor.

[0015] Preferably, the detected gas passes through the gas sensor, changing the gas concentration into a weak current signal I_Sin, which is output as a positive signal or a negative signal depending on the sensor principle; the I_Sin signal is input into the first polarity conversion unit U2-1 and the second polarity conversion unit U2-2, and the intelligent neural network calculation unit U5 automatically determines the sensor type and controls Sin_CTL1 and Sin_CTL2 to automatically switch the polarity. When I_Sin is a negative signal, the first analog switch S1 of the first polarity conversion unit U2-1 is turned on, connecting the signal I_Sin to R1B, and the second analog switch S2 of the second polarity conversion unit U2-2 is turned on; when I_Sin is a positive signal, the first analog switch S1 of the second polarity conversion unit U2-2 is turned on, connecting the signal I_Sin to R2B, and S2 of the first polarity conversion unit U2-1 is turned on.

[0016] Preferably, the gas sensor types include photoionization gas sensors, semiconductor gas sensors, contact combustion gas sensors, electrochemical gas sensors and non-dispersive infrared gas sensors, and the measurement range includes 0~10ppm, 10~100ppm, 100~1000ppm, 1000~10000ppm, greater than 10000ppm and 0~100%LEL.

[0017] The present invention also provides a method for calculating parameters of an adaptive system for multiple gas sensors, which is applied to the above-mentioned adaptive system for multiple gas sensors and includes the following steps:

[0018] Step 1: The intelligent neural network calculation unit U5 collects input information, including gas sensor type, measuring range, and manufacturer information, and digitizes the input information;

[0019] Step 2: The trained intelligent neural network operation unit U5 outputs the sensor coefficient vector C through forward propagation based on the input information. The sensor coefficient vector C includes the sensor polarity parameter Sin_PN, the sensor first-level gain parameter Ku3, the second-level gain parameter S_AMP, the second-order filter system function parameter, and the concentration zero voltage reference factor AE. 20 , concentration sensitivity parameter M 20 , cross-interference coefficients CF1, CF2…CF n and temperature correction factor rt;

[0020] Step 3: According to the sensor polarity parameter Sin_PN given by the intelligent neural network operation unit U5, control Sin_U2-1_CT1 and Sin_U2-2_CT1 to achieve automatic adaptation of the sensor polarity; the current signal I_Sin of the sensor is connected to the U1A circuit to realize the conversion of current and voltage, and the output voltage of the transconductance amplifier U1A is: , where Vref is the zero-adjusting voltage and R3 is the resistance value of the third resistor R3, converting the output current I_Sin of the gas sensor into a voltage signal;

[0021] Step 4: According to the first-stage gain parameter Ku3 of the gas sensor deduced by the intelligent neural network operation unit U5, the intelligent neural network calculation unit U5 automatically sets the resistance value of the digital resistor U3 ;

[0022] Step 5: According to the second-order filter system function parameter and the second-stage amplification gain weight parameter S_AMP of the filter of the gas sensor deduced by the intelligent neural network operation unit U5, the intelligent neural network operation unit U5 performs signal filtering and signal secondary amplification processing on the sensor;

[0023] Step 6: According to the concentration zero-voltage reference factor AE 20 , concentration sensitivity parameter M 20 , cross-interference coefficients CF1, CF2…CFn and temperature correction factor rt given by the intelligent neural network operation unit U5, use the above parameters for the calculation of the sensor concentration.

[0024] Preferably, the resistance value of the digital resistor U3 is set in the step 4 , and the formula is as follows:

[0025]

[0026] In the formula, Ku3 is the first-stage gain parameter of the sensor, R4 is the resistance value of the fourth resistor R4, and R6 is the resistance value of the fourth resistor R6.

[0027] Preferably, the system transfer function of the second-order analog filter in the step 5 is as follows:

[0028] The corresponding digital-analog filter system function:

[0029]

[0030] In the formula, a is the gain coefficient, b is the second-order term coefficient, is a complex variable on the unit circle,

[0031] is the natural constant, j is the imaginary unit, and T is the sampling period.

[0032] Preferably, in step 6, for the sensor cross-interference in ambient air, interference suppression needs to be performed, and the gas concentration calculation formula is as follows:

[0033]

[0034] Among them, AE 20 : Zero output voltage reference factor of the sensor at room temperature of 20°C;

[0035] VE: Real-time voltage of the sensor;

[0036] CFn: Cross-interference coefficient of the nth gas to the target gas;

[0037] Rair_n: Concentration of the nth gas;

[0038] rt: Temperature correction coefficient;

[0039] M 20 : Sensitivity parameter of the target gas at room temperature of 20°C.

[0040] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0041] 1. It adapts to gas sensors of different manufacturers, different principles, and different ranges, and realizes intelligent adaptation of multi-gas detection without increasing the product cost;

[0042] 2. The intelligent neural network calculation unit automatically matches the signal polarity of the gas sensor, automatically switches the range, and performs analog signal filtering and automatic signal gain matching, and can calculate the accurate concentration for different gases;

[0043] 3. According to the actual detection requirements, different types of gas sensors can be connected, and multiple gas sensors can be plugged and used on-site. Without increasing the hardware cost and without replacing the instrument equipment, the present invention realizes multi-purpose use of one instrument. Brief Description of the Drawings

[0044] Figure 1 is the system block diagram of the embodiment of the present invention;

[0045] Figure 2 is the system circuit schematic diagram of the embodiment of the present invention;

[0046] Figure 3 is Figure 2 partial enlarged view of the front-end polarity adaptation unit in

[0047] Figure 4 is Figure 2Partial enlarged view of the digital resistor U3, analog-to-digital conversion unit U4, and intelligent neural network calculation unit U5;

[0048] Figure 5 is Figure 2 Partial enlarged view of the display unit U6 in

[0049] Figure 6 Is the flow chart of intelligent neural network training in the embodiment;

[0050] Figure 7 Intelligent neural network calculation diagram in the embodiment. Specific implementation manner

[0051] The following further clarifies the present invention in conjunction with specific embodiments. The embodiments are implemented on the premise of the technical solution of the present invention. It should be understood that these embodiments are only used to illustrate the present invention and not to limit the scope of the present invention.

[0052] As Figures 1-5 shown, an adaptive system for multiple gas sensors includes a front-end polarity adaptation unit, a current-voltage conversion unit, an analog-to-digital conversion unit U4, an intelligent neural network calculation unit U5, a display unit U6, an alarm unit U7, and a remote terminal unit U8. The front-end polarity adaptation unit is used to connect to the gas sensor and receive the gas sensor current signal. The front-end polarity adaptation unit includes a first polarity transformation unit U2-1 and a second polarity transformation unit U2-2. The two polarity transformation units adopt existing digital switches, such as CMOS analog switches of the LTC52 series. The current-voltage conversion unit includes a transconductance amplifier U1A, a preamplifier U1B, and a digital resistor U3. The transconductance amplifier U1A is connected to the front-end polarity adaptation unit. The positive input terminal of the transconductance amplifier U1A is connected to the 2nd pin of the second polarity transformation unit U2-2 through the second resistor R2B. The negative input terminal of the transconductance amplifier U1A is connected to the 2nd pin of the first polarity transformation unit U2-1 through the first resistor R1B. The output terminal of the transconductance amplifier U1A is connected to the positive input terminal of the preamplifier U1B through the third resistor R3. The third resistor R3 is a transconductance resistor. The gas sensor current signal is converted into a voltage signal through the third resistor R3 and is primarily amplified by the preamplifier U1B. The negative input terminal of the preamplifier U1B is connected to the fourth resistor R4, and the negative input terminal of the preamplifier U1B is also connected to the digital resistor U3 through the sixth resistor R6. The output terminal of the preamplifier U1B is connected to the intelligent neural network calculation unit U5. The intelligent neural network calculation unit U5 uses a large-scale logic array FPGA chip. The intelligent neural network calculation unit U5 includes a first part unit and an intelligent parameter operation unit. The first part unit ( Figure 2 and Figure 4The Part1) in it is a filter and amplifier (Filter + AMP) designed by FPGA. The intelligent parameter operation unit (AI-Para_CompU) is an intelligent parameter operation unit designed by FPGA. The digital resistor U3 is connected to the intelligent neural network computing unit U5 through the I 2 2C buses SCL0 and SDA0. The intelligent neural network computing unit U5 automatically calculates the gain of the primary amplification according to the type, range, and manufacturer information of the gas sensor, and controls the output resistance of the digital resistor U3. The output terminal of the preamplifier U1B is connected to the first part unit of the intelligent neural network computing unit U5. The analog-to-digital conversion unit U4 is connected to the intelligent neural network computing unit U5. The intelligent neural network computing unit U5 controls the analog-to-digital conversion unit U4 to perform the conversion between analog and digital quantities to obtain the detected gas voltage value of the gas sensor. The intelligent neural network computing unit U5 automatically calculates the sensitivity parameter, sensor environment calibration parameter, and zero point parameter of the gas sensor according to the gas sensor type, range, and manufacturer information, that is, the gas sensor parameter Air_D, and finally calculates the current gas concentration. The display unit U6 is used to display the gas detection concentration display and alarm display. The alarm unit U7 is used for over-limit early warning. The remote terminal unit U8 is used for remote data interaction, obtains the gas sensor type, range, and manufacturer information, and synchronizes the detected gas concentration and alarm information to the remote data center.

[0053] In the detection site, the gas to be detected passes through a gas sensor, and the gas concentration is converted into a weak current signal I_Sin. According to different sensor principles, this signal outputs a positive or negative signal. The I_Sin signal is input into the first polarity conversion unit U2-1 and the second polarity conversion unit U2-2. The intelligent neural network calculation unit U5 automatically determines the sensor type and controls Sin_CTL1 and Sin_CTL2 (these two are polarity control signals, Sin_CTL1 is connected to pin 6 of the first polarity conversion unit U2-1, and Sin_CTL2 is connected to pin 6 of the second polarity conversion unit U2-2) to perform automatic polarity switching. Inside the first polarity conversion unit U2-1 and the second polarity conversion unit U2-2, there are respectively a first analog switch S1 and a second analog switch S2. When I_Sin is a negative signal, the first analog switch S1 of the first polarity conversion unit U2-1 conducts, connecting the signal I_Sin to R1B, and the second analog switch S2 of the second polarity conversion unit U2-2 conducts. When I_Sin is a positive signal, the first analog switch S1 of the second polarity conversion unit U2-2 conducts, connecting the signal I_Sin to R2B, and S2 of the first polarity conversion unit U2-1 conducts. I_Sin passes through the transconductance amplifier U1A for transconductance operation, is converted into a voltage signal through the third resistor R3, and then, through the third resistor R3, is connected to the preamplifier U1B for primary amplification. The amplification gain = (digital resistor U3 + R6) / R4, where the digital resistor U3 is controlled by the I 2 2C bus SCL0 and SDA0. The intelligent neural network calculation unit U5 uses a large-scale logic circuit FPGA to automatically calculate the gain of primary amplification based on sensor characteristics such as sensor type, range, and manufacturer information, and controls the output resistance of the digital resistor U3 through the I 2 2C bus. The voltage signal SOut output by the preamplifier U1B is sent to the first part unit (Part1) Filter+AMP of the intelligent neural network calculation unit U5 for digital filtering and secondary amplification to meet the input requirements of the analog-to-digital conversion unit U4. The parameters of the Filter in the Part1 unit are automatically calculated by the intelligent neural network calculation unit U5 according to the type and range of the gas sensor, and different gas sensor parameters can be input. The intelligent neural network calculation unit U5 establishes a neural network self-learning knowledge base to automatically match the relevant parameters of the gas sensor.

[0054] The signal Sin_AD, which has been filtered and amplified by the first part of the intelligent neural network computing unit U5, namely Filter + AMP, is connected to the analog-to-digital conversion unit U4. Under the timing control of the intelligent neural network computing unit U5, the analog-to-digital conversion unit U4 converts Sin_AD into a digital quantity VE. The intelligent neural network computing unit U5 uses the gas sensor matching parameters in the experience library for Air_D (gas sensor type, range, and manufacturer information), such as the gas sensor zero output voltage reference factor, sensitivity parameter, temperature correction coefficient, cross-interference coefficient, etc., and finally converts it into the gas concentration. Then it judges the gas concentration. The intelligent neural network computing unit U5 controls the alarm unit U7 to give an automatic early warning when it exceeds the standard, and the intelligent neural network computing unit U5 controls the display unit U6 to display the detected concentration and alarm display, etc. The intelligent neural network computing unit U5 conducts remote data interaction through the remote terminal unit U8, obtains the gas sensor parameters Air_D, including information such as gas sensor type, range, and manufacturer information, and synchronizes the detected gas concentration, alarm information, etc. to the remote data center.

[0055] The gas sensor types include photoionization gas sensors (PID), semiconductor gas sensors, catalytic combustion gas sensors, electrochemical gas sensors, and non-dispersive infrared gas sensors (NDIR), etc. Figure 2 and Figure 3 in, 3 / 4Electrode PID NDIR … sensor represents different types of gas sensors. 3 / 4Electrode means that three-electrode sensors and four-electrode sensors belong to existing electrochemical gas sensors. PID represents photoionization gas sensors, and NDIR represents non-dispersive infrared gas sensors.

[0056] The range includes 0~10ppm, 10~100ppm, 100~1000ppm, 1000~10000ppm, greater than 10000ppm, and 0~100%LEL.

[0057] The manufacturer information includes brand manufacturers such as CITY, Alpha, Honey, and SUSA. The manufacturer information represents the principle and polarity requirements of the sensor, and can be expanded later.

[0058] The intelligent neural network computing unit U5 is a self-learning network model based on the backpropagation (BP) algorithm. It is designed as a multi-layer feedforward neural network, which is trained through sensor input feature parameters and finally outputs the desired output coefficients required for subsequent calculations. As Figure 6 shown, drawing on the neural network BP algorithm and using gradient search technology, the intelligent neural network of the intelligent neural network computing unit U5 achieves the effect of self-learning through the training process.

[0059] The model of the intelligent neural network computing unit U5 is divided into an input layer, a hidden layer, and an output layer;

[0060] (1) The data of the input layer is the feature vector X = [x1, x2, x3] of the parameter Air_D (gas sensor parameter), and the number of nodes is the number of parameters n = 3 of Air_D;

[0061] The gas sensor parameters include the gas sensor type (principle factor), the measurement range (range factor), and the manufacturer information (brand factor);

[0062] The gas sensor types include photoionization gas sensors (PID), semiconductor gas sensors, catalytic combustion gas sensors, electrochemical gas sensors, and non-dispersive infrared gas sensors (NDIR), etc.;

[0063] The measurement ranges include 0~10ppm, 10~100ppm, 100~1000ppm, 1000~10000ppm, greater than 10000ppm, and 0-100%LEL;

[0064] The manufacturer information includes brand manufacturers such as CITY, Alpha, SUSA, Honey, etc., and the manufacturer information represents the principle and polarity requirements of the sensor.

[0065] (2) The number of layers of the hidden layer is set to 6, which is set according to the actual task complexity to avoid overfitting. The number of nodes i in this layer is set to twice the number of input parameters, that is, i = 6, and the function Softmax is used as the activation function.

[0066] (3) The number of nodes m in the output layer is consistent with the dimension of the subsequent calculation coefficients, and the coefficient vector C = [c1, c2,... c m .

[0067] The subsequent calculation coefficient vector C of the sensor includes: sensor polarity parameter, sensor first-stage gain parameter, second-stage gain parameter, filtering coefficient, concentration zero voltage reference factor, concentration sensitivity parameter, cross-interference coefficient, and temperature correction coefficient.

[0068] The process of the BP learning algorithm is as follows:

[0069] Step 1. Forward propagation

[0070] Step 1.1. The input layer receives the gas sensor feature vector X;

[0071] Step 1.2. The hidden layer calculates the activation values layer by layer:

[0072] h k = f(W k ﹒hk-1 +b k )

[0073] Among them, h k-1 is the activation value of each layer, W k is the weight matrix, b k is the bias, and f is the activation function;

[0074] Step 1.3: Obtain the weights and biases from the hidden layer to the output layer, and the output layer generates the coefficient vector C;

[0075] Step 2: Backpropagation

[0076] Step 2.1: Loss function: Mean Squared Error (MSE) or other task-related losses:

[0077]

[0078] Among them, is the true coefficient label, is the coefficient generated by the output layer, m is the number of output coefficients, is the mean squared error;

[0079] Step 2.2: Gradient calculation:

[0080] Calculate the weight gradients of each layer through the chain rule and bias gradients ;

[0081] Update the parameters using the gradient descent method:

[0082]

[0083] Among them, is the learning rate, is the weight gradient, is the bias gradient, and b k is the bias.

[0084] The self-learning mechanism process is as follows:

[0085] (1) Online learning

[0086] Real-time update: During the deployment phase, continuously update the network parameters, including the activation values of each layer in the hidden layer and the coefficients generated by the output layer, according to the newly collected sensor parameter Air_D and the corresponding true coefficient label .

[0087] Incremental training: Only use 1 - 2 new data each time to avoid global retraining.

[0088] (2) Adaptive adjustment

[0089] Dynamic learning rate: The learning rate is automatically adjusted using the Adam optimizer;

[0090] Regularization: The Dropout strategy is added to prevent overfitting.

[0091] The trained intelligent neural network computing unit U5 obtains the optimal weight gradients of each layer and bias gradients . The intelligent neural network computing unit U5 can then calculate the subsequent calculation coefficient vector C of the sensor based on the input sensor characteristic parameters for the automatic parameter calculation of the gas sensor.

[0092] As Figure 7 shown, this embodiment also discloses an adaptive parameter calculation method for multiple gas sensors, which is applied to the above-mentioned adaptive system for multiple gas sensors and includes the following steps:

[0093] Step 1, data preprocessing. The intelligent neural network computing unit U5 collects the Air_D (gas sensor parameters) of the input sensor to be replaced, and the information is sent to the intelligent neural network computing unit U5 through the serial port. The input information includes the gas sensor type, range, and manufacturer information, and the input information is digitally aggregated; the text information is converted into digital quantities for data aggregation processing, and the input characteristic parameters are concentrated into the numbers [0, 1, 2, 3, 4, 5] for the input of subsequent algorithms; and the training set, validation set, and test set are divided multiple times. The sensor parameters correspond to the numbers as follows:

[0094] (1) Gas sensor type (Air_D.Type):

[0095] 0: Photoionization gas sensor; 1: Non-dispersive infrared gas sensor; 2: Electrochemical gas sensor; 3: Semiconductor; 4: Catalytic combustion gas sensor;

[0096] (2) Range (Air_D.Range):

[0097] 0: 0~10 ppm; 1: 10~100 ppm; 2: 100~1000 ppm; 3: 1000~10000 ppm;

[0098] 4: Greater than 10000 ppm; 5: 0-100% LEL;

[0099] (3) Manufacturer information (Air_D.Factory): Four brand manufacturers, CITY, Alpha, Honey, and SUSA. The aggregation method is to assign digital quantities, CITY: 0; Alpha: 1; Honey: 2; SUSA: 3.

[0100] Step 2: Training phase. The intelligent neural network operation unit U5 assigns the sensor feature parameters obtained by digital aggregation in Step 1 to the sensor feature vector X, and inputs the feature vector X into the intelligent neural network U5 through the communication port to obtain the network prediction coefficients , and perform algorithm training:

[0101] Step 2.1: Input the sensor feature vector X → network prediction coefficients → Calculate the loss L → Update the parameter W through backpropagation k .

[0102] Step 2.2: Stop training according to the validation set loss.

[0103] Step 3: Inference phase. Input the feature vector X of the new sensor parameters, and through forward propagation, output the sensor coefficient vector C, including the sensor polarity parameter Sin_PN, the sensor first-stage gain parameter Ku3, the second-stage gain parameter S_AMP, the second-order filter system function parameters (a, b), the concentration zero point AE 20 , the concentration sensitivity parameter M 20 , the cross-interference coefficient CF n and the temperature correction coefficient rt.

[0104] Step 4: Set the sensor polarity change unit

[0105] According to the sensor polarity parameter Sin_PN given by the intelligent neural network operation unit U5, control Sin_U2-1_CT1 and Sin_U2-2_CT1 to achieve automatic adaptation of the sensor polarity. The current signal I_Sin of the sensor is connected to the U1A circuit to realize the conversion of current and voltage. The output voltage of U1A: , where Vref is the zero-adjusting voltage, and R3 is the resistance value of the third resistor R3. The third resistor R3 converts the output current I_Sin of the gas sensor into a voltage signal.

[0106] Step 5: According to the first-stage gain parameter Ku3 of the gas sensor deduced by the intelligent neural network operation unit U5, the intelligent neural network calculation unit U5 automatically sets the resistance value of the digital resistor U3 , and the formula is as follows:

[0107]

[0108] In the formula, R4 is the resistance value of the fourth resistor R4, and R6 is the resistance value of the fourth resistor R6;

[0109] Step 6: Based on the second-order filter system function parameters (including the gain coefficient a and the second-order term coefficient b) of the filter Filter of the gas sensor calculated by the intelligent neural network operation unit U5 and the secondary amplification gain weight parameter S_AMP, the intelligent neural network calculation unit U5 automatically performs signal filtering and signal secondary amplification processing on the sensor;

[0110] The system transfer function of the second-order analog filter is as follows:

[0111]

[0112] In the formula, is the transfer function, is the complex variable in the complex plane, a is the gain coefficient, and b is the second-order term coefficient;

[0113] The corresponding digital-analog filter system function:

[0114]

[0115] In the formula, is the system function, is the complex variable on the unit circle, is the natural constant, j is the imaginary unit, a is the gain coefficient, b is the second-order term coefficient, and T is the sampling period.

[0116] Step 7: The concentration zero voltage reference factor AE 20 , sensitivity parameter M 20 , cross-interference coefficients CF1, CF2...CFn, and temperature correction coefficient rt of the gas sensor calculated by the intelligent neural network operation unit U5 are used by the intelligent neural network calculation unit U5 for the calculation of the sensor concentration;

[0117] For the cross-interference of the sensor in ambient air, interference suppression is required. The gas concentration calculation formula is as follows:

[0118]

[0119] Among them, AE 20 : The zero-output voltage reference factor of the sensor at room temperature of 20°C;

[0120] VE: The real-time voltage of the sensor;

[0121] CFn: The cross-interference coefficient of the nth gas to the target gas;

[0122] Rair_n: The concentration of the nth gas (data obtained in advance);

[0123] rt: Temperature correction coefficient;

[0124] M 20 : Sensitivity parameter of the target gas at room temperature of 20°C.

[0125] Taking a certain gas sensor as an example:

[0126] The H2S sensor of Honeywell (Honey), with a measurement range of 0 - 10 ppm, is an electrochemical sensor. It is connected to the system of this embodiment, and the manufacturer information, measurement range, and sensor type are input into the device through the serial port. The parameters such as the polarity, gain, filtering coefficient, and concentration calculation coefficient of the sensor are calculated by the intelligent neural network operation unit U5, and finally the corresponding gas concentration is obtained through the sensor current value.

[0127] Step 1: Extract the characteristic parameter Air_D of the sensor through data preprocessing:

[0128] Air_D = {2, 0, 2}; / / Air_D.Ttype = 2; Air_D.Range = 0; Air_D.Factory = 2.

[0129] Step 2: Training stage (completed in the early stage).

[0130] Step 3: Inference stage

[0131] Input the sensor parameter feature vector X (X0 = Air_D.Ttype, X1 = Air_D.Range, X2 = Air_D.Factory):

[0132] X = {2; 0; 2};

[0133] Obtain the sensor coefficient vector C = {

[0134] Sin_PN = {Sin_U2 - 1_CT1 = 1, Sin_U2 - 2_CT1 = 0}; / / Positive polarity signal

[0135] Ku3 = 20;

[0136] S_AMP = 50;

[0137] a = {1, -1.1429, 0.4128};

[0138] b = {0.0674, 0.1349, 0.06745};

[0139] AE 20 = 250;

[0140] M 20 = 0.9;

[0141] CF = {0.0005, 0.2}; / / Interference coefficient of H2S for CO and SO2

[0142] rt = 0.9; / / 0.85 - 1.10

[0143] }。

[0144] Step 4: Setting of the sensor polarity change unit. The output of the sensor is a positive - polarity signal, and the positive - input terminal of U1A should be used.

[0145] Sin_U2 - 1_CT1 = 1;

[0146] Sin_U2 - 2_CT1 = 0.

[0147] Step 5: Set the resistance value of the digital potentiometer U3 ;

[0148] According to the formula,

[0149]

[0150] In the formula, ,

[0151] 。

[0152] Step 6: Set the digital filter and the signal secondary gain:

[0153] Set the filter according to parameters a and b;

[0154] Set the secondary gain coefficient of the signal according to S_AMP = 5.

[0155] Step 7:

[0156] Use a 1 - ppm standard gas and introduce it into the H2S sensor. At this time, the output current of the sensor I_Sin = 0.85 uA;

[0157] Let the resistance ,

[0158]

[0159] Then VE = 433.mV;

[0160] Concentration of the cross - interference gas CO (data measured by other sensors); Concentration of the cross - interference gas SO2 (data measured by other sensors);

[0161] Substitute the data into the following formula:

[0162]

[0163] M_Con (ppm) = 1.025 ppm.

[0164] Error from the theoretical data: 2.5%, meeting the design requirement of ±10%.

[0165] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. An adaptive system for multiple gas sensors, characterized in that, It includes a front-end polarity adaptation unit for connecting to a gas sensor and receiving a gas sensor current signal, a current-voltage conversion unit connected to the front-end polarity adaptation unit, an intelligent neural network calculation unit U5 connected to the current-voltage conversion unit, and an analog-digital conversion unit U4 connected to the intelligent neural network calculation unit U5. The intelligent neural network calculation unit U5 realizes signal front-end polarity matching by controlling the front-end polarity adaptation unit. The intelligent neural network calculation unit U5 converts the gas sensor current signal into a voltage signal by controlling the current-voltage conversion unit. The signal after front-end polarity matching and current-voltage conversion is filtered and amplified in signal gain by the intelligent neural network calculation unit U5, and then sent to the analog-digital conversion unit U4 after processing. The intelligent neural network calculation unit U5 controls the analog-digital conversion unit U4 to perform the conversion between analog quantity and digital quantity, obtains the detected gas voltage value of the gas sensor, and the intelligent neural network calculation unit U5 calculates the current gas concentration;The front-end polarity adaptation unit includes a first polarity transformation unit U2-1 and a second polarity transformation unit U2-2. The current-voltage conversion unit includes a transconductance amplifier U1A, a preamplifier U1B connected to the transconductance amplifier U1A, and a digital resistor U3 connected to the preamplifier U1B. The first polarity transformation unit U2-1 is connected to the transconductance amplifier U1A through a first resistor R1B, and the second polarity transformation unit U2-2 is connected to the transconductance amplifier U1A through a second resistor R2B. The intelligent neural network calculation unit U5 adapts to the type of gas sensor, activates and switches the first polarity transformation unit U2-1 and the second polarity transformation unit U2-2, and controls the analog switch polarity matching of the front-end polarity adaptation unit. The gas sensor current signal is connected to the transconductance amplifier U1A through the first polarity transformation unit U2-1 or the second polarity transformation unit U2-2. The transconductance amplifier U1A is connected to the preamplifier U1B through a third resistor R3. The gas sensor current signal is converted into a voltage signal through the third resistor R3 and is primarily amplified by the preamplifier U1B. The inverting input terminal of the preamplifier U1B is connected to a fourth resistor R4 and is connected to the digital resistor U3 through a sixth resistor R6. The intelligent neural network calculation unit U5 automatically calculates the gain of the primary amplification based on the type, range, and manufacturer information of the gas sensor, and controls the output resistance of the digital resistor U3. The output terminal of the preamplifier U1B is connected to the intelligent neural network calculation unit U5. The voltage signal output by the preamplifier U1B undergoes digital filtering and secondary amplification by the intelligent neural network calculation unit U5 to meet the input requirements of the analog-to-digital conversion unit U4. The intelligent neural network calculation unit U5 controls the analog-to-digital conversion unit U4 to perform the conversion between analog and digital quantities, obtaining the detected gas voltage value of the gas sensor. The intelligent neural network calculation unit U5 automatically calculates the sensitivity parameter, sensor environment calibration parameter, and zero point parameter of the gas sensor according to the gas sensor type, range, and manufacturer information, and finally calculates the current gas concentration.

2. The adaptive system for multiple gas sensors according to claim 1, wherein It also includes a display unit U6, an alarm unit U7 and a remote terminal unit U8 connected to the intelligent neural network computing unit U5. The display unit U6 is used to display the gas detection concentration and alarm display, the alarm unit U7 is used for exceeding the standard warning, and the remote terminal unit U8 is used for remote data interaction, obtaining the gas sensor type, range and manufacturer information, and synchronizing the detected gas concentration and alarm information to the remote data center.

3. The adaptive system for a plurality of gas sensors according to claim 1, wherein The intelligent neural network calculation unit U5 includes a first part unit and an intelligent parameter calculation unit. The first part unit is used for filtering and amplifying the signal, and the intelligent parameter calculation unit is used for parameter calculation of the gas sensor.

4. The adaptive system for multiple gas sensors according to claim 1, wherein The detected gas passes through the gas sensor, changing the gas concentration into a weak current signal I_Sin. Depending on the sensor principle, the signal is output as a positive signal or a negative signal; the I_Sin signal is input into the first polarity conversion unit U2-1 and the second polarity conversion unit U2-2, and the intelligent neural network calculation unit U5 automatically determines the sensor type and controls Sin_CTL1 and Sin_CTL2 to automatically switch the polarity. When I_Sin is a negative signal, the first analog switch S1 of the first polarity conversion unit U2-1 is turned on, connecting the signal I_Sin to R1B, and the second analog switch S2 of the second polarity conversion unit U2-2 is turned on; when I_Sin is a positive signal, the first analog switch S1 of the second polarity conversion unit U2-2 is turned on, connecting the signal I_Sin to R2B, and S2 of the first polarity conversion unit U2-1 is turned on.

5. The adaptive system for a plurality of gas sensors according to claim 1, wherein The gas sensor types include photoionization gas sensors, semiconductor gas sensors, contact combustion gas sensors, electrochemical gas sensors and non-dispersive infrared gas sensors, and the measurement ranges include 0~10ppm, 10~100ppm, 1000~10000ppm, greater than 10000ppm and 0~100%LEL.

6. A method for calculating adaptive system parameters for multiple gas sensors, which is applied to the adaptive system for multiple gas sensors as described in any one of claims 1-5, and is characterized in that, The following steps are involved: Step 1: The intelligent neural network calculation unit U5 collects input information, including gas sensor type, measuring range, and manufacturer information, and digitizes the input information; Step 2: Based on the input information, the trained intelligent neural network operation unit U5 outputs the sensor coefficient vector C through forward propagation. The sensor coefficient vector C includes the sensor polarity parameter Sin_PN, the sensor primary gain parameter Ku 3 , the secondary gain parameter S_AMP, the second-order filter system function parameter, the concentration zero voltage reference factor AE 20 , the concentration sensitivity parameter M 20 , the cross-interference coefficients CF1, CF2…CF n and the temperature correction coefficient rt; Step 3. Based on the sensor polarity parameter Sin_PN given by the intelligent neural network operation unit U5, control Sin_U2-1_CT1 and Sin_U2-2_CT1 to achieve automatic adaptation of the sensor polarity; the sensor current signal I_Sin is connected to the U1A circuit to achieve current-to-voltage conversion. The output voltage of the transconductance amplifier U1A is: V_TCA_1 = Vref + R3*I_Sin, where Vref is the zero adjustment voltage and R3 is the resistance value of the third resistor R3. The output current I_Sin of the gas sensor is converted into a voltage signal; Step 4: Based on the first-stage gain parameter of the gas sensor calculated by the intelligent neural network operation unit U5 Ku 3. The intelligent neural network calculation unit U5 automatically sets the resistance value of the digital resistor U3 ; Step 5: According to the second-order filter system function parameters of the filter of the gas sensor and the secondary amplification gain weight parameter S_AMP calculated by the intelligent neural network operation unit U5, the intelligent neural network operation unit U5 performs signal filtering and signal secondary amplification processing on the sensor; Step 6: Based on the concentration zero voltage reference factor AE of the gas sensor given by the intelligent neural network operation unit U5 20 , the concentration sensitivity parameter M 20 , the cross-interference coefficients CF1, CF2…CFn and the temperature correction factor rt, use the above parameters for the calculation of the sensor concentration.

7. The adaptive system parameter calculation method for a multi-gas sensor according to claim 6, characterized in that, Set the resistance value of the digital resistor U3 in step 4 , and the formula is as follows: ； Wherein, Ku 3 is the first-stage gain parameter of the sensor, R4 is the resistance value of the fourth resistor R4, and R6 is the resistance value of the fourth resistor R6.

8. The adaptive system parameter calculation method for a plurality of gas sensors according to claim 6, characterized in that The system transfer function of the second-order analog filter in the said Step 5 is as follows: ； The corresponding digital-analog filter system function: ； In the formula, a is the gain coefficient, b is the second-order term coefficient, z is the complex variable on the unit circle, e^ is the natural constant, j is the imaginary unit, and T is the sampling period.

9. The adaptive system parameter calculation method for a plurality of gas sensors according to claim 6, wherein In the said Step 6, for the sensor cross-interference in ambient air, interference suppression needs to be performed, and the gas concentration calculation formula is as follows: ； Among them, AE 20 : Zero output voltage reference factor of the sensor at room temperature of 20°C; VE: The real-time voltage of the sensor; CFn: The cross-interference coefficient of the nth gas to the target gas; Rair_n: The concentration of the nth gas; rt: The temperature correction coefficient; M 20 : Sensitivity parameter of the target gas at room temperature of 20°C.

Images