An adaptive sampling flow dynamic allocation system and method

Abstract

The application discloses a kind of self-adapting sampling flow dynamic distribution system and method thereof, including sampling unit, detection cavity and embedded control unit, the gas of different monitoring point is collected by the sampling unit and is all converged to the detection cavity, sampling unit further includes first sampling branch, second sampling branch and third sampling branch, each sampling branch is connected with flowmeter and electric proportional regulating valve, detection cavity further includes pre-preprocessing cavity and multispectral smoke sensor, embedded control unit calculates the fire risk index of each sampling branch according to the data of sampling unit and the flow detection value of each flowmeter, and dynamically distributes sampling flow between multiple sampling branches based on the fire risk index of each sampling branch, embedded control unit closed-loop adjusts the rotating speed of constant-flow vacuum pump according to the detection value of total flow sensor.The application is conducive to improving the early-stage smoke detection sensitivity, reducing the false alarm and missed alarm rate, and realizing automatic positioning of fault point.

Description

Technical Field

[0001] The present invention relates to the technical field of fire safety, in particular to an adaptive sampling flow dynamic allocation system and method. Background Art

[0002] The semiconductor dry etching process relies on a radio frequency power supply to excite special process gas to form plasma to complete wafer etching processing. The radio frequency electrodes, radio frequency coils and radio frequency matching devices in the radio frequency chamber of the etching machine are core heat-generating high-risk components. In the initial stage of process abnormality, only ppb-level ultra-diluted pyrolysis smoke is generated around local components. The air exchange frequency in semiconductor clean rooms is as high as 30~60 times per hour, and the extremely strong air circulation will rapidly dilute and diffuse trace smoke. In addition, the radio frequency working condition of the etching machine will generate 50~200Hz wide-band periodic electromagnetic interference, which is very easy to couple with the weak smoke signal, so it is difficult to capture the extremely early fire signal only by fixed-point fixed-flow sampling. The aspirating smoke detection systems matched with existing etching machines generally adopt a fixed flow equalization sampling structure preset at the factory or manually adjusted, the suction flow of each sampling branch is constant, and cannot be dynamically adjusted according to real-time working conditions and fire risk.

[0003] There have been several improvements in the existing industry technologies. For example, the multi-directional smoke inlet photoelectric smoke detector provided by Chinese Patent Publication No. CN210222980U improves the night detection accuracy through a detachable LED light ring; in the field of aspirating smoke detectors, there are also improvements that use eddy current dampers and dynamic flow controllers to stabilize the sampling airflow velocity, and use multiple detection modules to perform independent parallel detection to reduce the false alarm rate. However, none of the above solutions realizes competitive dynamic allocation of sampling flow among multiple sampling branches based on the fire risk index of each branch, and the fixed flow structure has the following deficiencies: First, the sampling resource allocation is rigid and unbalanced, the sampling flow of high-risk points is the same as that of low-risk points, and the limited detection resources cannot be focused on high-risk heat-generating components; second, the extremely early trace smoke signal is very easy to be diluted and submerged and covered by electromagnetic interference due to insufficient fixed flow of the corresponding branch, resulting in missing alarms and missing the disposal opportunity; third, the system can only identify the overall smoke concentration abnormality, cannot distinguish the risk level of each branch and locate the fault point, and requires manual point-by-point investigation, which is time-consuming and poor in accuracy.

[0004] Therefore, it is necessary to develop an adaptive sampling flow dynamic allocation system and method that overcomes the deficiencies of the prior art. Contents of the Invention

[0005] The present invention provides an adaptive sampling flow dynamic allocation system and method, which solves the problems that the existing aspirating smoke detection system for etching machines adopts a fixed flow equalization sampling structure, cannot dynamically allocate sampling resources according to real-time working conditions and fire risks, resulting in extremely early trace smoke signals being easily diluted and submerged, fault point positioning being delayed, and the missing alarm and false alarm rates being relatively high.

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: In a first aspect, the present invention provides an adaptive sampling flow rate dynamic allocation system, comprising a sampling unit, a total flow sensor, a detection chamber, an embedded control unit, and a constant flow vacuum pump. The sampling unit collects gas from different monitoring points and all flows into the detection chamber. The sampling unit further comprises a first sampling branch, a second sampling branch, and a third sampling branch. Each of the first, second, and third sampling branches is connected to a flow meter and an electric proportional regulating valve. The signal output terminal of the flow meter and the control terminal of the electric proportional regulating valve are both connected to the embedded control unit. The detection chamber further comprises a pre-processing chamber and a multispectral smoke sensor. The embedded control unit calculates the fire risk index of each sampling branch based on the data from the sampling unit and the flow detection values ​​of each flow meter, and dynamically allocates the sampling flow rate among the multiple sampling branches based on the fire risk index of each sampling branch. The total flow sensor is installed on the main pipeline after the sampling unit aggregates the flow. The embedded control unit adjusts the speed of the constant flow vacuum pump in a closed loop based on the detection value of the total flow sensor.

[0007] Furthermore, the embedded control unit calculates the unit flow concentration contribution factor of each branch based on the ratio of the smoke response signal of each sampling branch to the flow meter detection value of the corresponding branch, and calculates the fire risk index of each branch based on the weighted summation of the mean, concentration change rate and trend stability coefficient of the unit flow concentration contribution factor within the sliding window.

[0008] Furthermore, the embedded control unit normalizes the fire risk index of each sampling branch after index mapping to obtain the flow distribution weight of each branch, and controls the opening degree of the electric proportional regulating valve of each branch according to the flow distribution weight.

[0009] Furthermore, when the fire risk index of any sampling branch of the sampling unit exceeds a preset threshold for a preset number of consecutive sampling periods, is the highest globally for a preset number of consecutive sampling periods, and the difference between the fire risk index of the sampling branch and the second highest branch exceeds a preset difference threshold for a preset number of consecutive sampling periods, the embedded control unit locks the current flow allocation state and outputs the fault location signal of the sampling branch.

[0010] Furthermore, the pre-treatment chamber includes a cyclone separation section, a filter membrane section, and a condensate drainage section connected in sequence. The multispectral smoke sensor is a dual-wavelength sensor, with a blue light band of 430~470nm for detecting organic pyrolysis smoke and an infrared band of 900~950nm for avoiding interference from cleanroom water vapor.

[0011] Furthermore, the embedded control unit further includes a signal conditioning circuit subunit, an analog-to-digital conversion circuit subunit, a hardware risk index calculation circuit subunit, a hardware flow competition allocation circuit subunit, and a first PWM drive circuit subunit. The signal conditioning circuit subunit amplifies, low-pass filters, and performs power frequency notch filtering on the outputs of the multispectral smoke sensor and the flow meter. The analog-to-digital conversion circuit subunit digitizes the data at a sampling rate greater than or equal to 1 kHz. The hardware risk index calculation circuit subunit calculates the fire risk index. The hardware flow competition allocation circuit subunit converts the fire risk index into flow allocation weights. The first PWM drive circuit subunit drives the opening of the electric proportional control valve according to the flow allocation weights.

[0012] Furthermore, the embedded control unit also includes a hardware watchdog circuit subunit, a second PWM drive circuit subunit, and a sampling subunit. The sampling subunit communicates with the multispectral smoke sensor via an SPI bus. The hardware watchdog circuit subunit is connected to an SPI bus timeout detection circuit and a signal range comparator. When a communication interruption or sensor output exceeding the range is detected, the second PWM drive circuit subunit forces all electric proportional regulating valves to switch to the equal flow state.

[0013] Furthermore, the hardware flow contention allocation circuit subunit also includes an adaptive operating condition adjustment coefficient lookup table. The address of the adaptive operating condition adjustment coefficient lookup table is formed by concatenating the cleanroom ISO level code and the etching machine RF power quantization level, and outputs the adaptive operating condition adjustment coefficient. The index mapping is calculated by dividing the fire risk index of each branch by... The quantized value is then input.

[0014] In a second aspect, the present invention provides an adaptive sampling traffic dynamic allocation method, executed by the adaptive sampling traffic dynamic allocation system described in the first aspect, comprising: Step S1: Hardware signal preprocessing and feature extraction. The smoke response signal and flow signal of each sampling branch are collected synchronously at a sampling frequency of greater than or equal to 1 kHz. The unit flow concentration contribution factor is calculated by the hardware divider and stored in the sliding window. Step S2: Hardware multi-level risk level calculation, based on sliding window data to calculate the concentration change rate, trend stability coefficient and fire risk index of each sampling branch; Step S3: Hardware adaptive flow contention allocation. The fire risk index of each sampling branch is converted into flow allocation weight after index mapping and normalization. The opening of the electric proportional regulating valve is controlled by PWM to dynamically allocate the sampling flow. Step S4: Hardware confidence lock and early warning output. When the fire risk index of a certain sampling branch continuously meets the conditions of exceeding the threshold, global maximum and difference exceeding the threshold, the flow allocation status is locked and the fault location signal of the branch is output.

[0015] The beneficial effects of this invention are as follows: By independently setting flow meters and flow regulating valves for each sampling branch and using an embedded control unit to competitively and dynamically allocate sampling flow among branches based on the fire risk index of each branch, the limited total sampling flow can be focused on branches with high fire risk, which is beneficial to improving the detection sensitivity of very early trace smoke and reducing the false alarm rate; by using the embedded control unit to complete the risk index calculation and flow allocation through deterministic hardware logic, it is beneficial to ensure the real-time performance and reliability of the control under the strong electromagnetic interference of the etching machine; by locking the flow allocation state and outputting the fault point signal when the risk index of a certain branch continuously meets the preset conditions, it is beneficial to realize the automatic location of the fault point and reduce manual point-by-point inspection; by using the total flow sensor and constant flow vacuum pump to form a closed-loop control of the total flow, the total sampling flow of the system is kept constant when the valve opening of each branch changes dynamically, providing an accurate premise for competitive flow allocation; by using a hardware watchdog circuit to forcibly switch to the safe state of equal flow allocation when communication is interrupted or the sensor fails, it is beneficial to ensure that the basic monitoring function of the system does not fail. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the hardware architecture of the adaptive sampling flow dynamic allocation system of the present invention; Figure 2 This is a schematic diagram of the circuit composition of the embedded control unit of the present invention; Figure 3 This is a measured curve of the dynamic operating condition response of the hardware traffic contention allocation circuit of the present invention; Figure 4 This is a flowchart of the adaptive sampling flow dynamic allocation method of the present invention.

[0017] Explanation of reference numerals in the attached drawings: Sampling unit 11, first sampling branch 111, second sampling branch 112, third sampling branch 113, total flow sensor 12, detection cavity 13, pre-processing cavity 131, multispectral smoke sensor 132, embedded control unit 14, constant current vacuum pump 15, sampling subunit 1401, hardware watchdog circuit subunit 1402, second PWM drive circuit subunit 1403, period counter 1404, multiplexer 1405, hardware latch 1406, branch encoding circuit 1407, flow meter 1408, signal conditioning circuit subunit 1409, analog-to-digital conversion circuit subunit 1410, hardware risk index calculation circuit subunit 1411, hardware flow contention allocation circuit subunit 1412, first PWM drive circuit subunit 1413, electric proportional regulating valve 1414. Detailed Implementation

[0018] Example 1: System Hardware Architecture and Sampling Flow Contention Allocation Principle refer to Figure 1 The present invention provides a schematic diagram of the hardware architecture of an adaptive sampling flow dynamic allocation system. The system specifically includes a sampling unit 11, a total flow sensor 12, a detection cavity 13, an embedded control unit 14, and a constant flow vacuum pump 15. The sampling unit 11 further includes a first sampling branch 111, a second sampling branch 112, and a third sampling branch 113, corresponding to the three high-risk heat-generating components of the etching machine: the RF electrode, the RF coil, and the RF matching unit. Further reference is provided. Figure 2 The circuit diagram of the embedded control unit shows that the air passages of the first sampling branch 111, the second sampling branch 112, and the third sampling branch 113 are connected in series with the flow meter 1408 and the electric proportional regulating valve 1414 along the airflow direction. The signal output terminal of the flow meter 1408 is connected to the embedded control unit 14, and the control terminal of the electric proportional regulating valve 1414 is electrically connected to the PWM output terminal of the embedded control unit 14. The flow of each branch is independently controlled by the embedded control unit 14, and the hardware response time of the valve body is no more than 50ms. The air outlets of each sampling branch are collected into the detection chamber 13.

[0019] The detection chamber 13 further includes a pre-treatment chamber 131 and a multispectral smoke sensor 132. The pre-treatment chamber 131 is sequentially configured with a cyclone separation section, a filter membrane section, and a condensate drainage section: the cyclone separation section uses centrifugal force to remove large dust particles from the sample gas; the filter membrane section uses a polytetrafluoroethylene microporous filter membrane to filter fine particulate matter; the condensate drainage section uses a semiconductor cooling chip to lower the sample gas temperature below the dew point, condensing water vapor and discharging it through a steam trap. The pre-treated sample gas is then sent to the multispectral smoke sensor 132, which is a dual-wavelength sensor. The blue light band (430-470nm) is used for detecting organic pyrolysis smoke, and the infrared band (900-950nm) is used to avoid interference from cleanroom water vapor. The signal ratio of the two bands is used to distinguish between real smoke and non-fire interference such as water vapor and dust.

[0020] The total sampling flow rate of the system is provided by the constant flow vacuum pump 15, with a rated flow rate of 10L / min. The total flow sensor 12 is installed on the main pipeline after the sampling unit 11. The embedded control unit 14 adjusts the speed of the constant flow vacuum pump 15 through incremental PID closed loop according to its detection value, so that the total sampling flow rate of the system is maintained within the range of 10L / min±5%. When the opening degree of each branch valve changes and causes the main pipeline impedance to change, the flow fluctuation is automatically compensated. The proportional coefficient, integral coefficient, and derivative coefficient are calibrated according to the pump body characteristics and pipeline impedance.

[0021] The embedded control unit 14 further includes auxiliary circuits and core circuits. The core circuits include a hardware risk index calculation circuit subunit 1411, a hardware flow contention allocation circuit subunit 1412, a cycle counter 1404, a multiplexer 1405, a hardware latch 1406, and a branch encoding circuit 1407. The auxiliary circuits include a sampling subunit 1401, a hardware watchdog circuit subunit 1402, a second PWM drive circuit subunit 1403, a signal conditioning circuit subunit 1409, an analog-to-digital conversion circuit subunit 1410, and a first PWM drive circuit subunit 1413.

[0022] The sampling subunit 1401 reads the dual-band scattered light intensity data of the multispectral smoke sensor 132 via the SPI bus. The input terminal of the signal conditioning circuit subunit 1409 is connected to the output terminals of the multispectral smoke sensor 132 and each branch flow meter 1408. It includes an instrumentation amplifier, a second-order active low-pass filter (cutoff frequency 5Hz, used for anti-aliasing) and a 50Hz power frequency notch circuit. The second-order active low-pass filter limits the bandwidth of the 1kHz sampled analog signal to within 5Hz to prevent aliasing. The 50Hz power frequency notch circuit can handle the cutoff frequency shift caused by filter element aging or temperature drift, as well as the strong power frequency interference introduced by the cleanroom power supply system. The analog-to-digital conversion circuit subunit 1410 uses a 16-bit ADC to digitize the conditioned analog signal at a sampling rate of greater than or equal to 1kHz. The 1kHz sampling rate ensures that there are enough sampling points in the 50~200Hz interference frequency band for digital notch filtering and adaptive filtering, while providing sufficient anti-aliasing margin for downsampling to 10Hz.

[0023] The input of the hardware risk index calculation circuit subunit 1411 is connected to the output of the analog-to-digital conversion circuit subunit 1410. Internally, it integrates a hardware divider, a sliding window storage array, a hardware cascaded linear regression operation circuit, a discreteness calculation circuit, and a hardware weighted summator. The hardware divider performs a division operation between the smoke response voltage signal and the corresponding branch flow signal, outputting a unit flow concentration contribution factor and storing it in the sliding window storage array. The sliding window storage array is a 50×16-bit dual-port RAM, using an incremental update method. Each new sampling point written triggers the incremental calculation of the risk index, without waiting for the full 5-second window to fill. During system startup or when the window is not full, the total number of sampling points in the sliding window is taken as the currently stored actual number of sampling points. Once the window is full, it is always 50. The hardware cascaded linear regression operation circuit calculates the concentration change rate based on the sliding window data. The discreteness calculation circuit outputs a trend stability coefficient based on the standard deviation and mean of the concentration contribution factor within the sliding window. The hardware weighted summator outputs the fire risk index based on the weighted sum of the mean of the concentration contribution factor, the concentration change rate, and the trend stability coefficient.

[0024] The input of the hardware flow contention allocation circuit subunit 1412 is connected to the output of the hardware risk index calculation circuit subunit 1411. Internally, it integrates an adaptive operating condition adjustment coefficient calculation circuit, an exponential calculation ROM lookup table, a normalization divider, and a PWM mapping circuit. The adaptive operating condition adjustment coefficient calculation circuit is a hardware lookup table. The lookup table address is a 5-bit address formed by concatenating the cleanroom ISO level (2-bit code, covering ISO Class 3 to ISO Class 6) and the etching machine RF power signal (quantized to a 3-bit level by a comparator, covering the 0% to 100% power range), with a total of 32 entries. The output is an 8-bit fixed-point number representing the adaptive operating condition adjustment coefficient τ. Higher ISO levels or higher RF power values ​​indicate higher levels of adjustment. The smaller the value, the more pronounced the difference in traffic proportion among high-risk branches. The index calculation ROM lookup table divides the risk index of each branch by... The quantized value is input as an address and outputs a fixed-point approximation of exp(x). The normalization divider normalizes the exponent values ​​of each branch to flow distribution weights. The PWM mapping circuit linearly maps the weights to PWM duty cycle register values. The PWM frequency is 1kHz. All operations are performed using fixed-point scaling to avoid floating-point operation overhead.

[0025] The output of the hardware flow contention allocation circuit subunit 1412 is connected to the control terminal of the electric proportional regulating valve 1414 of each sampling branch via the first PWM drive circuit subunit 1413. The input terminals of the period counter 1404 and the multiplexer 1405 are connected to the output terminal of the hardware risk index calculation circuit subunit 1411. The input terminal of the hardware latch 1406 is connected to the output terminal of the multiplexer 1405. The output terminal of the hardware latch 1406 is connected to the input terminal of the branch encoding circuit 1407 and the enable terminal of the hardware flow contention allocation circuit subunit 1412. When a certain branch... When the road risk index exceeds a preset threshold for N consecutive sampling periods, is the highest globally for N consecutive sampling periods, and the difference between it and the second highest branch exceeds a preset difference threshold for N consecutive sampling periods, the hardware latch 1406 locks the current flow allocation state, and the branch encoding circuit 1407 outputs a switch signal corresponding to the fault point. The branch encoding circuit 1407 is a 3-bit binary encoder that outputs a switch signal corresponding to the fault branch number. Each circuit subunit is synchronized with a unified 1kHz system clock, the data path has a bit width of 16 bits, and pipeline registers are set in key operation nodes to ensure timing convergence.

[0026] The hardware watchdog circuit subunit 1402 is connected to an SPI bus timeout detection circuit and a signal range comparator. The SPI bus timeout detection circuit monitors the SPI communication between the sampling subunit 1401 and the multispectral smoke sensor 132. If no response is received from the sensor after a preset timeout period, the communication is considered interrupted. The signal range comparator monitors the digital signal output by the ADC. If the sensor output voltage is lower than 0.1V or higher than 90% of the full scale, the sensor is considered to have failed. The fault signal triggers the safety logic input of the hardware watchdog circuit subunit 1402, which forces all branch electric proportional regulating valves 1414 to switch to the even flow distribution state via the second PWM drive circuit subunit 1403, ensuring that the basic monitoring function does not fail.

[0027] The symbols used in this embodiment are defined as follows: c is the concentration contribution factor symbol, i is the sampling point or branch index, j is the branch index, ti is the timestamp of the i-th sampling point (in seconds), ci is the unit flow concentration contribution factor of the i-th sampling point (dimensionless, obtained by dividing the smoke response voltage by the corresponding branch flow), and n is the total number of sampling points in the sliding window (constantly 50 after the window is filled). The mean of the concentration contribution factor within the window is given by k, the concentration change rate (unit 1 / s), σ is the standard deviation within the window, s is the trend stability coefficient (dimensionless), R is the fire risk index (dimensionless), Ri is the fire risk index of the i-th branch, and α, β, and γ are the hardware fixed weight coefficients (α=0.1, β=0.6, γ=0.3). Here, ei is the adaptive operating condition adjustment coefficient (dimensionless), ej is the index value of the i-th branch, ej is the index value of the j-th branch in the summation operation, wi is the flow allocation weight of the i-th branch, Qi is the allocated flow of the i-th branch (unit L / min), and Qtotal is the total sampled flow of the system (10L / min).

[0028] The mathematical functions are defined as follows: exp is the natural exponential function, max is the maximum value function, and Σ is the summation operation. The fire risk index is calculated using the following formula: concentration change rate k = (n·Σ(ti·ci) -Σti·Σci) / (n·Σti² -(Σti)²), where k=0 when it is negative; trend stability coefficient s = max(0, 1-σ / ),when When the concentration is below the preset threshold of 0.01, s = 0; Fire risk index R = max(0, α· + β·k + γ·s), the competitive allocation of sampling flow for each branch is given by the following formula: the exponential value ei of the i-th branch is exp(Ri / The weight of the flow allocation for the i-th branch is wi = ei / Σ(ej), and the flow allocation for the i-th branch is Qi = Qtotal·wi. This is implemented using a hardware lookup table. The lookup table address is a 5-bit address concatenated from the cleanroom ISO level (2-bit code) and the etching machine RF power level (3-bit code), with a total of 32 entries. The output is an 8-bit fixed-point number. Higher ISO levels or higher RF power levels result in higher values. The smaller the value, the more it amplifies the difference in traffic share among high-risk branches. All the above values ​​are calculated using fixed-point numbers.

[0029] The system hardware fixed parameters are defined as follows: hardware sampling frequency not less than 1kHz, output feature value at 10Hz after hardware downsampling filtering, hardware sliding time window of 5s (50 sampling points), fixed weights of α=0.1, β=0.6, γ=0.3 under normal working conditions, normal judgment threshold of 0.7, continuous judgment for 20 sampling cycles, high risk judgment threshold of 0.85, continuous judgment for 10 sampling cycles, difference threshold of 0.1, and preset concentration threshold of 0.01.

[0030] like Figure 3 The measured curve of the dynamic operating condition response of the hardware flow contention allocation circuit of this invention is shown. The measured curve of the dynamic operating condition response shows that: at t≈5s, a process abnormality occurs at the point corresponding to the first sampling branch 111, and its risk index begins to rise. The hardware contention allocation circuit synchronously calculates the weights, Q1 gradually increases, and Q2 and Q3 gradually decrease, while the total flow rate remains constant at 10L / min; at t≈8s, R1 continuously meets the locking condition, and the hardware latch is triggered. At this time, Q1 is about 8.0L / min, and Q2 and Q3 are each about 1.0L / min. The branch encoding circuit outputs the fault point signal, and the flow rate does not oscillate after locking. The above curves and comparative data are based on tests obtained from a self-developed analog semiconductor production line. The test conditions are a simulated ISO Class 4 clean room, full-power radio frequency interference, and ultra-diluted smoke of 0.01~0.5mg / m³ to simulate an early fire. A total of 120 abnormal injection experiments were completed. Under the same test conditions, the comparison results between this system and the fixed flow rate distribution system are as follows: Risk update time for trace smoke below 0.1 mg / m³: this system is no more than 3 seconds, while the fixed flow rate distribution system is no less than 10 seconds; Fire detection accuracy: this system is no less than 99%, while the fixed flow rate distribution system is approximately 92%; False alarm rate: this system is 0.5%, while the fixed flow rate distribution system is approximately 5%; Fault location: this system automatically outputs the location, while the fixed flow rate distribution system requires manual inspection and the location time is no less than 30 minutes; End-to-end hardware control loop delay: this system is no more than 200 ms, the pure hardware solution has a control loop delay jitter of ±5 ms and runs continuously for 72 hours without crashes, while the software solution has a scheduling jitter of approximately ±50 ms and experiences an average of one program crash or watchdog reset every 8 hours under full-power RF interference. The above comparative data are based on simulated production line test conditions and do not extend to the full operating conditions of an actual production line.

[0031] Example 2: refer to Figure 4 The flowchart of the adaptive sampling flow dynamic allocation method of this invention shows that after the device is powered on, the embedded control unit 14 loads the fixed hardware parameters, and each electric proportional regulating valve 1414 distributes the flow evenly by default (approximately 3.33 L / min per channel), entering the real-time monitoring state. The specific steps include: Step S1: Hardware signal preprocessing and feature extraction. Specifically, the sampling subunit 1401 synchronously collects smoke and flow signals from each branch at a frequency of not less than 1kHz. After instrument amplification by the signal conditioning circuit subunit 1409, second-order active low-pass filtering (cutoff frequency 5Hz) and 50Hz power frequency notch filtering are used to eliminate electromagnetic interference. The signal is then digitized by the analog-to-digital conversion circuit subunit 1410 with 16-bit precision. The hardware divider divides the smoke response voltage signal with the corresponding branch flow signal to generate a unit flow concentration contribution factor. After hardware downsampling, the feature value is output at a frequency of 10Hz and cached in the sliding window storage array. The sliding window adopts an incremental update method, and the risk index incremental calculation is triggered after each new sampling point is written.

[0032] Step S2: Hardware multi-level risk level calculation, specifically: the hardware risk index calculation circuit subunit 1411 calculates the concentration change rate k based on sliding window data through a hardware cascaded linear regression calculation circuit, and calculates the standard deviation σ and mean of the concentration contribution factor within the window through a dispersion calculation circuit. The system outputs a trend stability coefficient s, which is then used by a hardware weighted summer to output the fire risk index R for each branch in real time. The risk update time is no more than 3 seconds and is not limited by the 5-second full window duration. Under normal operating conditions, R is approximately 0.001, far below the 0.7 threshold, and the system determines that there is no anomaly.

[0033] Step S3: Hardware adaptive flow contention allocation, specifically: when the risk index of a certain sampling branch increases, the hardware flow contention allocation circuit subunit 1412 combines the real-time read cleanroom ISO level and etching machine RF power signal, and... The adaptive operating condition adjustment coefficient is obtained by looking up the table. The flow weight of each branch is calculated by using the exponential operation ROM lookup table and the normalization divider. The result is converted into a drive level by the PWM mapping circuit. The opening degree of the electric proportional regulating valve 1414 of each branch is controlled by the first PWM drive circuit subunit 1413. The sampling resources of the low-risk branch are transferred to the high-risk branch. The valve body response time is no more than 50ms and the delay of the entire link hardware control loop is no more than 200ms.

[0034] Step S4: Hardware confidence locking and early warning output, specifically: when the risk index of a certain branch exceeds the corresponding threshold for N consecutive sampling periods, is the highest globally for N consecutive sampling periods, and the difference between it and the second highest branch exceeds 0.1 for N consecutive sampling periods, the cycle counter 1404 and the multiplexer 1405 output a locking signal, the hardware latch 1406 locks the current flow allocation state, stops dynamic adjustment to avoid adjustment oscillation, and at the same time, the branch encoding circuit 1407 outputs the corresponding fault point switch signal, triggering an audible and visual early warning and background prompt. Under normal operating conditions, N=20, threshold 0.7, and the early warning locking time is usually no more than 5s; under high-risk operating conditions, when the risk index exceeds 0.85, it automatically switches to the fast judgment channel with N=10 and threshold 0.85, and the early warning locking time is no more than 4s. The selection of the difference threshold of 0.1 is based on a safety margin of twice the typical value of the fluctuation amplitude of the risk index of each branch under normal operating conditions, which is less than 0.05, to avoid competitive oscillation when two branches are abnormal at the same time and the risks are close.

[0035] After troubleshooting, if the risk index of each branch is below the reset threshold (typically 0.3) for at least 5 seconds, the hardware latch 1406 automatically unlocks, the enable pin of the first PWM drive circuit subunit 1413 becomes active again, and the system returns to the dynamic flow contention allocation mode. The reset threshold and duration are designed to prevent adjustment oscillations caused by frequent switching during fault intervals. If an SPI communication interruption or sensor output out-of-range fault occurs, the hardware watchdog circuit subunit 1402 initiates safety logic, and the second PWM drive circuit subunit 1403 forces all branches to switch to the evenly distributed flow state (approximately 3.33 L / min per branch), ensuring that the basic monitoring function does not fail.

Claims

1. An adaptive sampling flow dynamic allocation system, characterized by, The system includes a sampling unit, a total flow sensor, a detection chamber, an embedded control unit, and a constant flow vacuum pump. The sampling unit collects gas from different monitoring points and converges it into the detection chamber. The sampling unit further includes a first sampling branch, a second sampling branch, and a third sampling branch. Each of the first, second, and third sampling branches is connected to a flow meter and an electric proportional control valve. The signal output terminal of the flow meter and the control terminal of the electric proportional control valve are both connected to the embedded control unit. The detection chamber further includes a pre-processing chamber and a multispectral smoke sensor. The embedded control unit calculates the fire risk index of each sampling branch based on the data from the sampling unit and the flow detection values ​​of each flow meter, and dynamically allocates the sampling flow rate among the multiple sampling branches based on the fire risk index of each sampling branch. The total flow sensor is installed on the main pipeline after the sampling unit converges the flow. The embedded control unit adjusts the speed of the constant flow vacuum pump in a closed loop based on the detection value of the total flow sensor.

2. The adaptive sampling flow dynamic allocation system of claim 1, wherein, The embedded control unit calculates the unit flow concentration contribution factor of each branch based on the ratio of the smoke response signal of each sampling branch to the flow meter detection value of the corresponding branch, and calculates the fire risk index of each branch based on the weighted sum of the mean of the unit flow concentration contribution factor, the concentration change rate and the trend stability coefficient within the sliding window.

3. The adaptive sampling flow dynamic allocation system of claim 1, wherein, The embedded control unit normalizes the fire risk index of each sampling branch after index mapping to obtain the flow distribution weight of each branch, and controls the opening degree of the electric proportional regulating valve of each branch according to the flow distribution weight.

4. The self-adapting sampling flow dynamic allocation system of claim 3, wherein, When the fire risk index of any sampling branch of the sampling unit exceeds a preset threshold for a preset number of consecutive sampling periods, is the highest globally for a preset number of consecutive sampling periods, and the difference between the fire risk index of the sampling branch and the second highest branch exceeds a preset difference threshold for a preset number of consecutive sampling periods, the embedded control unit locks the current flow allocation status and outputs the fault location signal of that sampling branch.

5. The adaptive sampling flow dynamic allocation system of claim 1, wherein, The pretreatment chamber includes a cyclone separation section, a filter membrane section, and a condensate drainage section connected in sequence. The multispectral smoke sensor is a dual-wavelength sensor, with a blue light band of 430~470nm for detecting organic pyrolysis smoke and an infrared band of 900~950nm for avoiding interference from water vapor in the clean room.

6. The self-adapting sampling flow dynamic allocation system of claim 4, wherein, The embedded control unit further includes a signal conditioning circuit subunit, an analog-to-digital conversion circuit subunit, a hardware risk index calculation circuit subunit, a hardware flow competition allocation circuit subunit, and a first PWM drive circuit subunit. The signal conditioning circuit subunit amplifies, low-pass filters, and applies power frequency notch filtering to the outputs of the multispectral smoke sensor and flow meter. The analog-to-digital conversion circuit subunit digitizes the data at a sampling rate greater than or equal to 1 kHz. The hardware risk index calculation circuit subunit calculates the fire risk index. The hardware flow competition allocation circuit subunit converts the fire risk index into flow allocation weights. The first PWM drive circuit subunit drives the opening of the electric proportional control valve according to the flow allocation weights.

7. The adaptive sampling flow dynamic allocation system according to claim 6, characterized in that, The embedded control unit also includes a hardware watchdog circuit subunit, a second PWM drive circuit subunit, and a sampling subunit. The sampling subunit communicates with the multispectral smoke sensor via an SPI bus. The hardware watchdog circuit subunit is connected to an SPI bus timeout detection circuit and a signal range comparator. When a communication interruption or sensor output exceeding the range is detected, the second PWM drive circuit subunit forces all electric proportional regulating valves to switch to the equal flow state.

8. The adaptive sampling flow dynamic allocation system according to claim 6, characterized in that, The hardware flow contention allocation circuit subunit also includes an adaptive operating condition adjustment coefficient lookup table. The address of the adaptive operating condition adjustment coefficient lookup table is formed by concatenating the cleanroom ISO level code and the etching machine RF power quantization level, and outputs the adaptive operating condition adjustment coefficient. The index mapping is calculated by dividing the fire risk index of each branch by... The quantized value is then input.

9. An adaptive sampling flow dynamic allocation method, characterized in that, Performed by the adaptive sampling flow dynamic allocation system of claim 1, comprising: Step S1: Hardware signal preprocessing and feature extraction. The smoke response signal and flow signal of each sampling branch are collected synchronously at a sampling frequency of greater than or equal to 1 kHz. The unit flow concentration contribution factor is calculated by the hardware divider and stored in the sliding window. Step S2: Hardware multi-level risk level calculation, based on sliding window data to calculate the concentration change rate, trend stability coefficient and fire risk index of each sampling branch; Step S3: Hardware adaptive flow contention allocation. The fire risk index of each sampling branch is converted into flow allocation weight after index mapping and normalization. The opening of the electric proportional regulating valve is controlled by PWM to dynamically allocate the sampling flow. Step S4: Hardware confidence lock and early warning output. When the fire risk index of a certain sampling branch continuously meets the conditions of exceeding the threshold, global maximum and difference exceeding the threshold, the flow allocation status is locked and the fault location signal of the branch is output.

Images