Artificial intelligence-based industrial dust removal energy-saving energy efficiency detection method and system

Abstract

The present application belongs to the technical field of industrial dust removal detection equipment, in particular to the technical field of detecting the concentration of suspended particles in the operating environment of industrial equipment, and more particularly to an industrial dust removal energy-saving energy efficiency detection method and system based on artificial intelligence. The present application uses the two independent parameters of current and pressure difference at the inlet and outlet of the dust collector to mutually verify and correct the dust concentration, so that the intelligent agent can verify the objectivity of the dust concentration monitored by the sensor through the other two parameters in time when the sensor has a problem, ensure that the overall system will not follow the wrong data, protect the energy-saving effect, and will not adjust the frequency randomly because of the bad sensor. Online all the way, low cost, only need to reuse the current, pressure difference and frequency data of the dust removal system. The system saves each round of data for model optimization, forms a virtuous cycle, and becomes more and more accurate.

Description

Technical Field

[0001] This invention belongs to the technical field of industrial dust removal testing equipment, specifically relating to the technical field of equipment for detecting the concentration of suspended particles in the operating environment of industrial equipment, and particularly to an artificial intelligence-based method and system for detecting energy-saving efficiency of industrial dust removal. Background Technology

[0002] In industries such as metallurgy, casting, and building materials, the production process generates a large amount of dust. Dust collectors are used to draw dust-laden gas into a dust collector, where filter bags trap the dust before the clean gas is discharged. The dust collector is the power source for the entire system, driven by a high-power motor (e.g., 110kW). Its speed can be adjusted by a frequency converter. Higher speed (higher frequency) results in larger air volume and stronger dust removal capacity, but also higher power consumption. Lower speed saves power, but may not remove dust completely, potentially leading to environmental degradation and penalties.

[0003] The traditional approach is to fix the fan frequency at a relatively high value (such as 48-50Hz) for a long time to cope with occasional dust peaks, at the cost of wasting electricity most of the time. In recent years, intelligent adjustment methods have emerged: allowing computers to automatically adjust the fan frequency according to the on-site dust concentration, increasing the frequency when there is a lot of dust to ensure dust removal effect, and decreasing the frequency when there is little dust to save energy.

[0004] However, in related technologies, the mainstream method for online continuous measurement of dust concentration in industrial dust collection pipelines is currently the light scattering sensor (the principle is similar to shining a beam of light into smoke and estimating the amount of dust by measuring the intensity of the scattered light). But it has several intractable drawbacks: the light-measuring lens can gradually become dirty with dust, leading to lower readings; it measures the intensity of scattered light rather than the weight of dust, and the conversion factor varies with the size, shape, and material of the dust particles; the concentration is uneven throughout the pipeline, and single-point sampling has limited representativeness. These drawbacks do not cause sudden malfunctions, but rather a slow shift in sensor readings. Once this shift occurs, the intelligent agent will make decisions based on a seemingly normal but actually low concentration, mistakenly believing there is little dust when there is actually a lot, and thus reducing the frequency, resulting in inadequate dust collection. In traditional systems, no link in the process can detect this error.

[0005] To address this kind of deviation, sensor self-calibration solutions (such as correction of readings based on contamination of the detection lens, dual-path compensation, and airflow protection lenses) only address the issue within the sensor itself; once a systemic misalignment occurs, it becomes difficult to detect with any internal self-calibration.

[0006] Therefore, there is an urgent need to design an industrial dust removal energy efficiency detection method and system based on artificial intelligence to solve the technical problem that when the sensor itself is contaminated or damaged, resulting in systemic inaccuracies, the system cannot detect it in time and continues to use the already inaccurate monitoring data, causing the fan power to fail to meet the energy-saving requirements.

[0007] It should be noted that the information disclosed in this background section is only for understanding the background technology of this application concept, and therefore may include information that does not constitute prior art. Summary of the Invention

[0008] This disclosure provides at least one method and system for detecting energy-saving efficiency in industrial dust removal based on artificial intelligence.

[0009] In a first aspect, embodiments of this disclosure provide an artificial intelligence-based method for detecting energy-saving efficiency in industrial dust removal, comprising the following steps: Step S1: Synchronously collect the operating current of the dust collector fan motor. Operating frequency Dust collector inlet pressure Export pressure And the measured value of dust concentration sensor And calculate the pressure difference dP; Step S2: Based on the physical correspondence between the fan motor current and the operating frequency, the current deviation component is obtained from the difference between the measured current and the expected current. Step S3: Based on the coupling relationship between pressure difference and dust concentration, construct a pressure difference-concentration consistency index and its deviation from the historical baseline. Step S4: Combine the current deviation component with the pressure difference-concentration consistency deviation to form a reliability score K for the dust concentration sensor reading; Step S5: Compare the credibility score K with the set threshold K th In contrast, when the confidence score is below the threshold, at least one of the following is performed on the sensor readings: deviation correction, downweighting, or alarm. Step S6: The long short-term memory network predicts the future dust concentration trend based on the historical concentration sequence; Step S7: The SAC reinforcement learning algorithm learns online and outputs the wind turbine frequency command, using the confidence-weighted state vector as the state, the inverter frequency as the action, and the reward consisting of specific energy consumption and emission exceeding penalties as the objective, and enters the next closed loop cycle.

[0010] In an optional implementation, in step S2, the expected current is determined by dividing the frequency into 2Hz increments when the fan is normal and the filter bag is clean, recording the average current of each increment, and accumulating data from at least three increments to complete the calibration; the expected current at any frequency is calculated using piecewise linear interpolation or quadratic polynomial fitting between increments. ; Among them, the two adjacent calibration points are a and b, and the frequency of point a is f a The current is I a The frequency of point b is f b The current is The frequency at any point is f ; The specific steps include: Step S21: Determine the gear position and frequency difference; Step S22: Determine the target frequency offset; Step S23: Determine the current difference; Step S24: Substitute the values ​​into the formula to calculate the ratio and add the intercept.

[0011] In an optional implementation, in step S3, the pressure difference-concentration consistency index is calculated according to the following formula: ; in, eps It is a constant; The deviation of the pressure-concentration ratio from the historical baseline is calculated using the following formula: , in, The historical baseline of system resistance corresponding to the measured unit dust concentration. What is the corresponding system resistance for a given unit dust concentration in real time? delta _ R The larger the value, the lower the confidence level of the sensor reading.

[0012] In an optional implementation, in step S4, the credibility score K is calculated using the following formula: ; in, , , It is the fusion weight of the three types of signals, and + + =1, norm() is a normalization process, clip(...,0,1) limits the result to between 0 and 1 to prevent it from going out of bounds; The closer the value is to 1, the more reliable the dust concentration reading is; The closer it is to 0, the less reliable it is.

[0013] In an alternative implementation, in step S5, the confidence score is... K With a standard preset threshold and low confidence threshold Differentiated and tiered handling based on credibility includes: Step S51, when K≥K th The sensor was deemed reliable and readings were normal, so it was directly... The original value is given to the decision-making process, and the wind turbine continues to adjust intelligently as usual; Step S52, ≤ < Mildly unreliable, downgraded in ranking, no forced correction; Step S53, < =0.3 Severely unreliable, correction, downgrading and alarm are executed simultaneously, the judgment is directly driven by the credibility score to determine the specific fault type; in, It is 0.3. It is 0.6.

[0014] In an optional implementation, in step S7, the confidence-weighted state vector is... i A vector of dimension s i Credibility is alpha i Confidence weight of dust concentration dimension alpha =K, the weighted state is

[0015] .

[0016] In one alternative implementation, the reward function r is: ; Where E represents energy consumption and M represents dust removal capacity. λ represents the emission limit, and λ is the penalty coefficient.

[0017] In an optional embodiment, the industrial dust removal energy efficiency testing method further includes a parameter discrimination step: based on the combination of current deviation, pressure difference-concentration consistency deviation and abnormal performance of concentration reading, the specific fault type is determined to be one of the following: the concentration sensor range is too low and drifting, the concentration sensor is interfered with or disconnected, the pressure difference sensor is abnormal or the sampling tube is blocked, or the actual load is increased.

[0018] Secondly, this disclosure also provides an artificial intelligence-based industrial dust removal energy efficiency testing system, including: The multi-source data acquisition module is used to synchronously acquire the current, operating frequency, inlet and outlet pressure difference of the dust collector, and measured values ​​of the dust concentration sensor of the main motor of the dust removal fan system. The reliability assessment module is used to calculate the current deviation component based on the physical correspondence between motor current and operating frequency, calculate the pressure difference-concentration consistency deviation based on the coupling relationship between pressure difference and dust concentration, and integrate the current deviation component and the pressure difference-concentration consistency deviation into a reliability score for the dust concentration sensor reading. The concentration prediction module, based on a long short-term memory network, is used to predict future dust concentration trends based on historical concentration sequences. The reinforcement learning decision-making module, based on the SAC algorithm, uses a confidence-weighted state vector as the state, the inverter frequency as the action, and the reward consisting of specific energy consumption and emission exceeding penalties as the objective. It learns online and outputs the wind turbine frequency command. The execution and feedback module is used to send frequency commands to the frequency converter for execution after safety checks, and to store the running data in the database for online model updates; And a module for performing at least one of deviation correction, deweighting or alarm on sensor readings when the confidence score is below a threshold, and feeding the confidence score back to the decision-making process.

[0019] In one alternative implementation, the multi-source data acquisition module further includes an assembly rack, which is mounted on the conveying pipe at the dust collector inlet. The assembly frame is separately equipped with a lens mounting base and a main body mounting base, which are coaxially arranged. The assembly frame is provided with a mating groove, and the lens mounting base can be detachably inserted into the mating groove; The assembly frame also has two guide rails, which are respectively set on both sides of the mating groove, and the main body mounting seat is slidably set between the two guide rails. The main body mounting seat is suitable for inserting a dust sensor. A limit cover plate is detachably installed on the side of the main body mounting seat away from the lens mounting seat. A light guide plate extends from the side of the main mounting base toward the lens mounting base, and the light guide plate is inserted into the lens mounting base. The inner wall of the end of the lens mounting base away from the main mounting base is provided with several slots along the circumference. The end of the lens mounting base near the main mounting base is provided with an annular groove. A positioning ring is provided at the end of the lens mounting base facing the main mounting base. The positioning ring is fitted with the annular groove. Several spring pieces are provided along the circumference of the positioning ring. The end of the spring piece away from the positioning ring is inserted into the slot. The spring is bent toward the center of the lens mounting base, and the minimum distance between the spring and the axis is less than the radius of the light guide plate. The spring is inclinedly disposed on the inner side wall of one end of the slot, and a positioning groove is provided on the inner side wall of the spring. A movable ring is also provided between the lens mounting base and the main body mounting base, and the movable ring is sleeved on the outer wall of the light guide plate; Several connecting springs are provided between the movable ring and the main mounting base, and the movable ring is adapted to be threaded to the side wall of the ring groove.

[0020] The beneficial effects of this invention are that it utilizes two independent parameters—current and the pressure difference between the inlet and outlet of the dust collector—to mutually verify and correct dust concentration. This allows the intelligent system to promptly verify the objectivity of the dust concentration detected by the sensor using these two additional parameters when the sensor malfunctions, ensuring that the overall system does not reuse erroneous data, guaranteeing energy efficiency, and preventing erratic frequency adjustments due to faulty sensors. It operates entirely online, is low-cost, and only requires reusing the current, pressure difference, and frequency data from the dust collection system. The system saves data from each round for model optimization, forming a virtuous cycle that improves accuracy with repeated use.

[0021] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained in accordance with the structures particularly pointed out in the description, claims and drawings.

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0023] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0024] Figure 1 A comparison diagram of a traditional intelligent agent and an optimized intelligent agent of the present invention is provided for embodiments of this disclosure. Figure 2 A flowchart illustrating a method for assessing the reliability of dust concentration measurements and correcting deviations, provided in this embodiment of the disclosure; Figure 3 A schematic diagram of the structure of an artificial intelligence-based industrial dust removal energy efficiency testing system provided in this embodiment of the disclosure; Figure 4 This is a schematic diagram of the assembly position of an assembly rack and a conveying pipe provided in an embodiment of the present disclosure; Figure 5An exploded view of an assembly rack and conveying pipeline provided in an embodiment of this disclosure; Figure 6 An exploded view of a lens mounting base and a main body mounting base provided in an embodiment of this disclosure; Figure 7 This is a schematic diagram of the structure of a lens mounting base provided in an embodiment of the present disclosure; Figure 8 This is a schematic diagram of the structure of a positioning ring provided in an embodiment of the present disclosure; Figure 9 This is a schematic diagram of the structure of a main mounting base provided in an embodiment of the present disclosure; Figure 10 This is a cross-sectional view of a lens mounting base and a main body mounting base in an assembled state, provided as an embodiment of this disclosure.

[0025] In the picture: 1. Transport pipeline; 2. Assembly rack; 21. Mating groove; 22. Guide rail; 3. Lens mounting base; 31. Slot; 32. Ring groove; 33. Positioning ring; 34. Spring piece; 35. Positioning groove; 4. Main mounting base; 41. Limiting cover plate; 42. Light guide plate; 43. Connecting spring; 5. Moving ring; 6. Isolation plate. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] Research has found that industries such as metallurgy, casting, and building materials generate large amounts of dust during production processes. Dust collectors are used to draw dust-laden gas into dust collectors, where filter bags trap the dust before clean gas is discharged. The dust collector is the power source for the entire system, driven by a high-power motor (e.g., 110kW). Its speed can be adjusted via a frequency converter. Higher speed (higher frequency) results in larger air volume and stronger dust removal capacity, but also higher power consumption; lower speed saves power, but may not remove all dust, potentially leading to environmental degradation and penalties.

[0028] Running the fan at a fixed high frequency (e.g., 48-50Hz) for extended periods to handle occasional dust spikes results in wasted electricity most of the time. A smarter approach, however, involves a computer that automatically adjusts the fan frequency based on the dust concentration. The frequency is increased when dust is high to ensure effective dust removal, and decreased when dust is low to save energy.

[0029] However, the aforementioned intelligent agent operates under a core assumption: the dust removal system follows a "perception-prediction-decision-execution-feedback" intelligent cycle. First, perception: multi-source sensing modules read on-site data; dust concentration sensors measure concentration, inlet and outlet pressure transmitters measure pressure difference, and current transformers measure the fan motor's operating current and frequency. Second, prediction: a Long Short-Term Memory (LSTM) network predicts short-term dust change trends based on historical concentration sequences. Third, decision: a reinforcement learning algorithm (SAC) calculates the optimal operating frequency of the fan using a weighted state vector based on credibility. Fourth, execution: the frequency command is sent to the frequency converter to adjust the fan speed. Fifth, feedback: after the fan speed changes, the on-site dust concentration changes accordingly, the sensors measure the new value, and the cycle begins again. In this way, the fan continuously adjusts itself automatically to follow dust changes, completing one cycle per rotation.

[0030] In this cycle, dust concentration is the most critical input, directly determining the agent's assessment of the current dust level and whether to increase or decrease the frequency. In other words, the overall decision-making quality of the agent rests on the assumption that the values ​​measured by the dust concentration sensor are accurate and reliable. However, this assumption often does not hold true in industrial settings, which is precisely the core problem this invention aims to solve.

[0031] Specifically, during dust concentration sensor detection, the lens can become gradually contaminated by dust or malfunction, causing readings to deviate. Once this deviation occurs, the agent will make decisions based on a seemingly normal but actually low concentration, mistakenly believing there is little dust when there is actually a lot, thus reducing the frequency and resulting in inadequate dust removal. In traditional systems, no step can detect this error. Traditional sensor self-calibration solutions, such as correcting readings based on lens contamination, dual-optical-path compensation, and airflow protection lenses, cannot prevent sensor malfunctions. Once a systemic inaccuracy occurs, no internal self-calibration can detect it.

[0032] In dust removal systems, besides the dust concentration parameters directly read from dust concentration sensors, two other parameters indirectly reflect the dust situation within the system: the pressure difference between the inlet and outlet of the dust collector and the current of the fan motor. Specifically, the greater the dust accumulation on the filter bags inside the dust collector, the greater the resistance and pressure difference; the denser the dust inside the dust collector, the greater the motor current. There is a clear physical relationship between these parameters and dust concentration: more dust → thicker dust accumulation on the filter bags → greater resistance and pressure difference → more effort required from the fan → greater current. In relevant AI-based dust removal solutions, it is generally assumed that the dust concentration sensor is accurate and objective, and no step in the entire cycle verifies the confidence level of the measured dust concentration.

[0033] The shortcomings of the above solutions are the result of the inventor's practical experience and careful research. Therefore, the discovery process of the above problems and the solutions proposed in this disclosure below should be considered as the inventor's contribution to this disclosure.

[0034] The following detailed description of some embodiments of the present invention is provided in conjunction with the accompanying drawings. Unless otherwise specified, the embodiments and features described below can be combined with each other. Furthermore, in the accompanying drawings, the thickness of components may be exaggerated or reduced for the purpose of effectively describing the technical content.

[0035] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0036] Based on the above research, and referring to Figure 2 This disclosure provides an artificial intelligence-based method for detecting energy-saving efficiency in industrial dust removal, comprising the following steps: Step S1: Multi-source data acquisition, synchronously acquiring the operating current of the dust collector fan motor. Operating frequency Dust collector inlet pressure Export pressure And the measured value of dust concentration sensor And calculate the pressure difference dP, dP= - ; Step S2: Based on the physical correspondence between the fan motor current and the operating frequency, the current deviation component is obtained from the difference between the measured current and the expected current. Step S3: Based on the coupling relationship between pressure difference and dust concentration, construct a pressure difference-concentration consistency index and its deviation from the historical baseline. Step S4: Combine the current deviation component with the pressure difference-concentration consistency deviation to form a reliability score K for the dust concentration sensor reading; Step S5: Compare the credibility score K with the set threshold K th In contrast, when the confidence score is below the threshold, at least one of the following is performed on the sensor readings: deviation correction, downweighting, or alarm. Step S6: The long short-term memory network predicts the future dust concentration trend based on the historical concentration sequence; Step S7: The SAC reinforcement learning algorithm, using a confidence-weighted state vector as the state, the inverter frequency as the action, and a reward consisting of penalties for exceeding energy consumption and emission limits as the objective, learns online and outputs the wind turbine frequency command, entering the next closed-loop cycle. SAC stands for SoftActor-Critic.

[0037] Reference Figure 1 Compared to traditional agent loops, referencing Figure 1 In this context, 'a' specifically refers to: Environment (controlled device) → Sensing (observing dust concentration), Feedback (real-time operating status) → Agent decision-making (adjusting frequency) → Environment (controlled device). In at least one embodiment, refer to... Figure 1 In step b, the agent loop is optimized and designed as follows: Environment (controlled device) → Perception (observation of dust concentration) → Credibility assessment → Agent reinforcement learning and weighted decision-making based on credibility, predicting future dust concentration trends based on feedback from the environment (controlled device) → Action (frequency adjustment) → Environment (controlled device). By utilizing the physical relationship of the dust collector fan and the coupling relationship between the inlet and outlet pressure difference and dust concentration, two types of physical verification signals independent of the dust concentration sensor are constructed. These signals are cross-compare with the measured values ​​of the dust concentration sensor to obtain a credibility score for the sensor reading. When the credibility score is lower than a threshold, at least one of the following is performed on the sensor reading: deviation correction, weight reduction, or alarm. The credibility score is then fed back to the decision-making process, ensuring the decision-maker remains robust even when the sensor malfunctions.

[0038] In at least one embodiment, in step S2, at the same fan speed (frequency), the more dust in the dust removal system, the greater the motor current. To further obtain the original current value at a certain frequency, i.e., the expected current, a self-learning calibration needs to be performed on-site. The frequency is divided into 2Hz increments, and the average current of each increment is recorded when the fan is normal and the filter bags are clean. Once data from at least three increments is accumulated, the calibration is complete. The values ​​between increments are calculated using a fitting formula. Thus, given any frequency, the expected current value at that frequency can be calculated, denoted as... The current deviation component is defined as: ; Among them, the two adjacent calibration points are a and b, and the frequency of point a is f a The current is Ia The frequency of point b is f b The current is The frequency at any point is f ; A specific calculation example is as follows: calibrating the average current at 28Hz. I a =84A, 32Hz average current I a =96A, then the expected current at 30Hz is calculated using the following steps: Step S21, determine the gear position and frequency difference: =28Hz, =32Hz, =32 28 = 4Hz.

[0039] Step S22, determine the target frequency offset: =30 28 = 2Hz.

[0040] Step S23, determine the current difference: =96 84 = 12A.

[0041] Step S24: Substitute the values ​​into the formula to calculate the ratio and add the intercept:

[0042] That is, the expected current at 30Hz is 90A.

[0043] Based on the above calculations, we can further calculate: ; This refers to the percentage by which the actual current exceeds the expected current. The larger the value, the heavier the system load.

[0044] In at least one embodiment, in step S3, the pressure difference-concentration consistency index is calculated according to the following formula:

[0045] in, eps Rdp is a constant, a very small number to avoid a denominator of 0. Its physical meaning is the system resistance corresponding to a unit dust concentration. When the filter bag is stable, this ratio should fluctuate slightly around a fixed value. However, if the dust concentration sensor is malfunctioning, it will report a different concentration. The value is too low, while the actual dust concentration is higher, increasing the pressure difference (dP). Therefore, the numerator becomes larger and the denominator becomes smaller. It will then spike abnormally. Its value was measured in a limited number of experiments during a period when the filter bag was clean and the concentration sensor had just been reliably calibrated, continuously recording the pressure difference (dP) and concentration for several minutes. The ratio is then averaged. It should be noted that recalibration is required if the operating conditions affecting this ratio change, such as: changing the fan model or overhauling the motor (changing the I / f relationship); significant changes in dust particle size or material (changing the resistance of the filter bag to the same mass of dust); replacement of the filter bag or after a thorough cleaning of the filter bag (changing resistance characteristics); or periodic automatic checks revealing changes. Under reliable operating conditions, the deviation from the baseline exceeds the set proportion continuously. That is, the initial measured calibration is combined with recalibration triggered by changes in conditions to ensure the universality of the method for different fan models and different dust conditions.

[0046] The deviation of the pressure-concentration ratio from the historical baseline is calculated using the following formula: , in, The historical baseline of system resistance corresponding to the measured unit dust concentration. What is the corresponding system resistance for a unit dust concentration in real time? delta _ R The larger the value, the lower the confidence level of the sensor reading.

[0047] In at least one embodiment, in step S4, the credibility score K is calculated using the following formula: ; in, , , It is the fusion weight of the three types of signals, and + + =1, norm() is a normalization process, clip(...,0,1) limits the result to between 0 and 1 to prevent it from going out of bounds; The closer the value is to 1, the more reliable the dust concentration reading is; The closer it is to 0, the less reliable it is.

[0048] In at least one embodiment, weights , , The determination is divided into two stages.

[0049] I. Initial Values ​​(Empirical Setting): Based on the principles of physical reliability, current and differential pressure are directly measured physical quantities, less prone to drift, and independent of the concentration sensor; therefore, higher weights are chosen. Reference concentration. Since the model is inversely derived and inherently contains errors, the weights are set to be relatively low. This embodiment uses empirical initial values. =0.4 (current) =0.4 (pressure difference) =0.2 (reference dust concentration), the sum of the three is 1.

[0050] 2. Offline optimization: After the system accumulates a certain amount of data with manual reference annotations, with the goal of maximizing the consistency rate between the credibility judgment and the manual references, Bayesian optimization is used to perform offline optimization on the three weights to obtain the optimal weights that are more suitable for the local equipment without changing the method structure.

[0051] Runtime weight redistribution principle: Let the confidence level of each dimension of the decision state vector be α. i (concentration dimension α) C = If the confidence level of the current / voltage difference dimension is always 1, then the reassigned weights are each α. i Divide by the sum of all α values ​​to ensure the sum is always 1. When the concentration confidence level K decreases, the concentration dimension automatically relinquishes its weight, while the current and pressure difference dimensions receive proportionally higher weights.

[0052] Threshold normalization is employed, and the specific executable algorithm is as follows. The purpose of normalization is to convert deviations of different dimensions into an abnormality level between 0 and 1. A tolerable threshold T is preset for each deviation, and deviations are divided by the threshold; deviations exceeding this threshold are truncated to 1. The formula is:

[0053] Each of the four items takes a threshold: current deviation =0.4, pressure difference deviation =0.3, concentration difference =1000mg / m³. The parameters nI, nR, and nC are calculated item by item. nI is the ratio of the current parameter deviation to the deviation threshold; nR is the ratio of the pressure difference-concentration consistency parameter deviation to the deviation threshold; and nC is the ratio of the dust concentration parameter deviation to the deviation threshold. The larger these three calculated values ​​are, the lower the reliability. ; (·,1) is truncated, and the result is directly substituted into the confidence weighted summation.

[0054] In step S5, the credibility score is... K With a standard preset threshold and low confidence threshold Differentiated and tiered handling based on credibility includes: Step S51, when K≥K th The sensor was deemed reliable and readings were normal, so it was directly... The original value is given to the decision-making process, and the wind turbine continues to adjust intelligently as usual; Step S52, ≤ < Mildly unreliable, downgraded in ranking, no forced correction; Step S53, < =0.3 Severely unreliable; correction, downgrading, and alarm are executed simultaneously. The judgment is directly driven by the reliability score to determine the specific fault type. in, It is 0.3. It is 0.6.

[0055] The correction is to use the concentration reference value. Lower readings Corrected, concentration reference value The constant is the one measured during normal system operation; the deweighting function tells the decision algorithm that the concentration has shifted, reducing the weight of the concentration in the decision-making process, while increasing the weight of the current and differential pressure, allowing the fan to make decisions based on more reliable signals; the alarm function notifies maintenance personnel to check and clean the sensors.

[0056] In at least one embodiment, when it is determined that correction is needed, linear fusion correction is used instead of directly discarding sensor readings. Instead, it compares the sensor readings with the concentration reference value. Weighted fusion based on credibility yields the corrected concentration. The formula is:

[0057] Among them, the fusion coefficient β is negatively correlated with the credibility, and β=1 is taken. ( The lower the value, the larger the β, and the greater the correction strength, limited to [0,1]:

[0058] This embodiment involves step-by-step calculations: (Known) =0.258、 =180mg / m³ =960mg / m³.

[0059] Step A1, calculate the fusion coefficient: β=1 0.258 = 0.742.

[0060] Step A2, calculate the weighted sum of the two paths:

[0061]

[0062] Step A3, add them together to get the corrected concentration:

[0063] Its effect is that the influence of the concentration signal automatically decays when the confidence level K decreases. It is directly driven by the confidence level score and adapts to the sensor status in real time without human intervention.

[0064] In step S7, after the SAC decision model calculates the wind turbine frequency command, it first undergoes a safety check (amplitude limit, variable rate limit) before being sent to the frequency converter for execution. Simultaneously, all data from this round (including confidence scores) is processed. Various deviations and predicted concentrations The final decision and reward (r) are stored as experience samples in the replay library for online updates to the SAC policy network and value network; the LSTM prediction model is also updated continuously with new concentration data. The longer the system is used and the more data it accumulates, the more accurate the predictions and decisions become, and the more significant the energy-saving effect.

[0065] In at least one embodiment, the industrial dust removal energy efficiency testing method further includes a parameter discrimination step: based on the abnormal combination of current deviation, pressure difference-concentration consistency deviation, and concentration reading, the specific fault type is determined as follows: 1. The current deviation is large and the pressure difference deviation is also large, but the concentration reading is stable and low, indicating that the concentration sensor range is low and drifting. 2. The concentration reading suddenly jumps around randomly, but the current and differential pressure are very stable. The concentration sensor is being interfered with or is disconnected. 3. The differential pressure deviation is large, but the current and concentration are stable, indicating that the differential pressure sensor is malfunctioning or the sampling tube is blocked; 4. The large current deviation, large pressure difference deviation, and simultaneous increase in concentration reading indicate an increase in the actual load.

[0066] In at least one embodiment, in step S6, the confidence-weighted state vector is... i A vector of dimension s i Credibility is alpha i Confidence weight of dust concentration dimension alpha =K, the weighted state is

[0067] .

[0068] In this way, when the dust concentration confidence level K is high, the concentration signal functions normally; when K is low, the influence of the concentration signal automatically weakens, and the model relies more on more reliable signals such as current and pressure difference to make judgments. Thus, even if the sensor malfunctions, the fan can still be adjusted properly, and the system will not collapse due to a faulty sensor. This mechanism makes the entire intelligent agent robust even when the sensor fails.

[0069] To ensure forward-looking decision-making, in at least one embodiment, a concentration prediction module based on a Long Short-Term Memory (LSTM) network is included after the confidence assessment and before the decision. This module takes the confidence-corrected concentration sequence of the most recent closed-loop cycles (e.g., 20 cycles) as input to predict the dust concentration trend for the next cycle (or several cycles). Let the historical concentration sequence be... (t 1) (t 2) ... (t n), the LSTM units sequentially read and update their hidden states, and finally the fully connected regression head outputs the predicted concentration for the next period. : ; The LSTM is a time-series regression network structure, whose parameters are trained offline and fine-tuned online using historical operational data. (Predicted concentration) With credibility score The various deviations together constitute the input of the decision-maker, enabling the fan to increase its frequency before dust rises significantly, thus avoiding dust removal delays.

[0070] In at least one embodiment, the frequency decision of the dust removal system is made by the SAC reinforcement learning algorithm. SAC is a mature off-policy deep reinforcement learning algorithm in the field, combining a policy network and a value network, and encouraging exploration through a maximum entropy regularization term, making it suitable for tasks such as dust removal that involve continuous actions and variable operating conditions. Its elements are defined as follows: State s: Consists of a state vector weighted by credibility, i.e. = × Each dimension includes confidence-corrected concentration and predicted concentration. Current deviation Pressure difference deviation wait.

[0071]

[0072] Action a: Continuous inverter frequency command f (e.g., value range 20~50Hz) is output by the strategy network.

[0073] Reward r: With the goal of minimizing energy consumption per unit of dust removal while ensuring emissions do not exceed standards, the reward function is defined as follows:

[0074] Where E is energy consumption, M is dust removal capacity, and the second item is the penalty for exceeding emission standards. Exceeding emission limits The system applies a penalty for exceeding the limit, with λ being the penalty coefficient. In the SAC online interaction, data (s, a, r, s′) is continuously collected, where s′ represents the state s at the next collection moment. The sample updates the policy network and value network. The longer the system runs and the more samples are collected, the more the decision tends towards the optimal energy-saving strategy.

[0075] In at least one embodiment, to facilitate intuitive understanding and application, a real-world application scenario is used to demonstrate how the calculation is performed. Taking a foundry electric furnace dust removal system as an example, during operation, the accuracy of the dust concentration sensor gradually deteriorates, reporting increasingly lower readings, while the actual dust concentration has significantly increased due to changes in the electric furnace's operating conditions. A traditional intelligent system would be misled by this consistently low reading, mistakenly assuming low dust levels and reducing its frequency, resulting in inadequate dust removal.

[0076] Based on this, a credibility threshold is set. =0.6, weights of the three parameters =0.4 (current) =0.4 (pressure difference) =0.2 (reference dust concentration). A comparison of data from normal and abnormal operating conditions is shown in Table 1:

[0077] Table 1 First, the current is abnormal. At the same 30Hz frequency, the current increased from 86A to 102A, indicating abnormal fan operation, increased dust, and current deviation.

[0078] Secondly, if the pressure difference is abnormal, calibrate the baseline R using normal operating conditions. dp，base =800 / (350+1)=2.279; This pressure difference-concentration ratio R dp =1200 / (180+1)=6.63, far exceeding the baseline, deviation.

[0079] Then, the concentration reference was abnormal, C_ref=960mg / m³, which differed from the sensor reading of 180 by |180|. 960|=780; Next, summarize the confidence levels: divide each of the three deviations by a threshold and normalize (current deviation threshold 0.4, pressure difference deviation threshold 0.3, concentration difference threshold 1000), resulting in 0.465, 1, and 0.78. After weighted summation, subtract 1 from the sum.

[0080] In summary, the confidence level K=0.258, far below the threshold of 0.6, indicates that the sensor reading is unreliable. The system immediately initiated fallback measures: correcting the low reading of 180 with a reference concentration of 960, while simultaneously reducing the weight of the concentration signal in the decision algorithm and switching to current and differential pressure to maintain a higher frequency to remove dust, and notifying maintenance personnel to clean the sensor. Even with sensor failure, the dust removal effect was preserved, preventing the agent from being misled and reducing its frequency.

[0081] Furthermore, this disclosure also provides an artificial intelligence-based industrial dust removal energy efficiency testing system, including: The multi-source data acquisition module is used to synchronously acquire the current, operating frequency, inlet and outlet pressure difference of the dust collector, and measured values ​​of the dust concentration sensor of the main motor of the dust removal fan system. The reliability assessment module is used to calculate the current deviation component based on the physical correspondence between motor current and operating frequency, calculate the pressure difference-concentration consistency deviation based on the coupling relationship between pressure difference and dust concentration, and integrate the current deviation component and the pressure difference-concentration consistency deviation into a reliability score for the dust concentration sensor reading. The concentration prediction module, based on a long short-term memory network, is used to predict future dust concentration trends based on historical concentration sequences. The reinforcement learning decision-making module, based on the SAC algorithm, uses a confidence-weighted state vector as the state, the inverter frequency as the action, and the reward consisting of specific energy consumption and emission exceeding penalties as the objective. It learns online and outputs the wind turbine frequency command. The execution and feedback module is used to send frequency commands to the frequency converter for execution after safety checks, and to store the running data in the database for online model updates; And a module for performing at least one of deviation correction, deweighting or alarm on sensor readings when the confidence score is below a threshold, and feeding the confidence score back to the decision-making process.

[0082] On the other hand, during the operation of the system, when one of the following occurs—correction, deweighting, or alarm—the dust sensor needs to be inspected on-site to meet further energy-saving requirements, based on the actual operating conditions. However, shutting down the dust removal system for the purpose of inspecting the dust sensor is also highly inappropriate.

[0083] Therefore, referring to Figure 3 In at least one embodiment, the multi-source data acquisition module further includes an assembly frame 2, which is installed on the conveying pipe 1 at the inlet of the dust collector. Since maintenance is performed without shutting down the dust collector, continuous operation of the dust collector can cause high-frequency vibration, affecting maintenance. Therefore, the conveying pipe 1 can be set separately from the dust collector. At the same time, the connection between the conveying pipe 1 and the dust collector can be made through a flexible connection, such as a corrugated pipe or other flexible hose, to offset the vibration impact during maintenance without shutting down the dust collector.

[0084] Reference Figure 3 and Figure 4 In at least one embodiment, the assembly frame 2 is separately equipped with a lens mounting base 3 and a main body mounting base 4, which are coaxially arranged. The lens mounting base 3 is used to mount a lens, and the main body mounting base 4 is used to mount the dust sensor body. The coaxial arrangement of the lens mounting base 3 and the main body mounting base 4 ensures unobstructed optical path between the lens and the dust sensor.

[0085] Reference Figure 5 In at least one embodiment, the assembly frame 2 has a mating groove 21, into which the lens mounting base 3 is detachably inserted. Simultaneously, the assembly frame 2 also has two guide rails 22, respectively disposed on both sides of the mating groove 21, and the main mounting base 4 is slidably disposed between the two guide rails 22, with a dust sensor inserted into the main mounting base 4. The lens mounting base 3 is positioned by the mating groove 21. For main mounting bases 4 designed with different models or parameters, radial positioning can be achieved first by using the guide rails 22, followed by axial positioning by installing pins on the guide rails 22. This allows for the initial positioning of the lens mounting base 3 and the main mounting base 4, as well as the positioning of their relative positions.

[0086] Reference Figure 9 In at least one embodiment, a limiting cover 41 is detachably mounted on the side of the main body mounting base 4 away from the lens mounting base 3. Optionally, the limiting cover 41 is threadedly connected to the main body mounting base 4, and the side of the limiting cover 41 facing the inside of the main body mounting base 4 has a spring. After the dust sensor is placed inside the main body mounting base 4, tightening the limiting cover 41 will cause the spring to press the dust sensor against the inner bottom wall of the main body mounting base 4, thereby achieving the positioning of the dust sensor.

[0087] Reference Figure 9 and Figure 10 In at least one embodiment, a light guide plate 42 extends from the side of the main body mounting base 4 toward the lens mounting base 3. The light guide plate 42 is inserted into the lens mounting base 3. The light guide plate 42 is annular and fits against the inner wall of the lens mounting base 3. By inserting the light guide plate 42 into the lens mounting base 3, the axis of the lens mounting base 3 and the main body mounting base 4 is ensured to coincide. At the same time, the light guide plate 42 is also a light guiding component connecting the light path between the lens mounting base 3 and the main body mounting base 4, ensuring that the light path is unobstructed.

[0088] Reference Figure 7 and Figure 8In at least one embodiment, the inner wall of the end of the lens mounting base 3 away from the main mounting base 4 is provided with a plurality of slots 31 along the circumferential direction, and the end of the lens mounting base 3 near the main mounting base 4 is provided with an annular groove 32. A positioning ring 33 is provided at the end of the lens mounting base 3 facing the main mounting base 4, and the positioning ring 33 is fitted into the annular groove 32. A plurality of spring pieces 34 are provided along the circumferential direction of the positioning ring 33, and the end of the spring piece 34 away from the positioning ring 33 is engaged in the slot 31. The spring piece 34 is inclined at one end of the inner wall of the slot 31, and a positioning groove 35 is provided on the inner wall of the spring piece 34. When installing the lens, align the lens with the side of the lens mounting base 3 with the slot 31. The end of the spring piece 34 protrudes from the lens mounting base 3. First, place the lens at the center of the end of each spring piece 34. The inclined surface of the spring piece 34 forms a funnel shape for initial positioning. Then, press the lens axially. The edge of the lens spreads the spring piece 34, causing it to deform outward until the lens reaches the positioning groove 35. At this point, each spring piece 34 elastically returns to its original position, securing the lens in place through the positioning groove 35. After the spring pieces 34 secure the lens, because they are evenly distributed circumferentially, they automatically align the lens with the center under the action of elastic force, coinciding with the axis of the lens mounting base 3.

[0089] Reference Figure 8 and Figure 10 In at least one embodiment, the spring 34 is bent toward the center of the lens mounting base 3, and the minimum distance of the spring 34 from the axis is less than the radius of the light guide plate 42. When disassembling the lens, the lens mounting base 3 is pulled out as a whole into the mating groove 21, so that the depth of the light guide plate 42 inserted into the lens mounting base 3 increases until the light guide plate 42 fits against the waist of each spring 34, which is the part where the spring 34 is bent to its maximum. Then, as the lens mounting base 3 continues to be pulled out, the light guide plate 42 pushes each spring 34 outward to open it, so that the lens is disengaged from the positioning groove 35, thereby realizing the installation of the lens.

[0090] Reference Figure 9 In at least one embodiment, a movable ring 5 is further provided between the lens mounting base 3 and the main body mounting base 4, and the movable ring 5 is sleeved on the outer wall of the light guide plate 42. A plurality of connecting springs 43 are provided between the movable ring 5 and the main body mounting base 4, and the movable ring 5 is adapted to be threadedly connected to the side wall of the annular groove 32. Through the threaded connection between the movable ring 5 and the annular groove 32, after the movable ring 5 is installed with the lens mounting base 3, the clamping force applied by the connecting springs 43 securely presses the lens mounting base 3 into the mating groove 21. On the other hand, since the positioning ring 33 is installed in the annular groove 32, when the movable ring 5 is stably connected to the annular groove 32 via threads, the movable ring 5 can also press the positioning ring 33, thereby achieving the positioning of each spring 34 and the lens.

[0091] Reference Figure 4In at least one embodiment, an isolation plate 6 is slidably disposed on one side of the assembly frame 2. The isolation plate 6 is disposed between the inner cavity of the conveying pipe 1 and the mating groove 21. The isolation plate 6 is used to isolate the passage between the conveying pipe 1 and the mating groove 21 when changing lenses without stopping the system, so as to avoid dust leakage. At the same time, the side of the isolation plate 6 facing the lens mounting seat 3 also has a brush. If the lens is only contaminated with some dust, causing the dust removal system to issue a correction or degrading warning, the isolation plate 6 can be closed first. By manipulating the movable ring 5, the lens mounting seat 3 can be rotated slightly to make the lens mounting seat 3 contact the brush and wipe the lens surface, in order to attempt to repair the accuracy of the lens, without having to replace the lens directly.

[0092] In some embodiments, after the lens is replaced or installed, the isolation plate 6 is opened, the dust sensor is reconnected to the system, and the above steps are repeated. If the reading returns to normal, the system continues to run and intelligently adjusts the motor frequency to achieve the best energy-saving effect. If it fails to return to normal, the dust sensor should be replaced.

[0093] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Based on the above-described ideal embodiments of the present invention, and through the above description, those skilled in the art can make various changes and modifications without departing from the technical spirit of the disclosed embodiments. The technical scope of the embodiments of this disclosure is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. An artificial intelligence-based method for detecting energy-saving efficiency in industrial dust removal, characterized in that, Includes the following steps: Step S1: Synchronously collect the operating current of the dust collector fan motor. Operating frequency Dust collector inlet pressure Export pressure And the measured value of dust concentration sensor And calculate the pressure difference dP; Step S2: Based on the physical correspondence between the fan motor current and the operating frequency, the current deviation component is obtained from the difference between the measured current and the expected current. Step S3: Based on the coupling relationship between pressure difference and dust concentration, construct a pressure difference-concentration consistency index and its deviation from the historical baseline. Step S4: Combine the current deviation component with the pressure difference-concentration consistency deviation to form a reliability score K for the dust concentration sensor reading; Step S5: Compare the credibility score K with the set threshold K th In contrast, when the confidence score is below the threshold, at least one of the following is performed on the sensor readings: deviation correction, downweighting, or alarm. Step S6: The long short-term memory network predicts the future dust concentration trend based on the historical concentration sequence; Step S7: The SAC reinforcement learning algorithm learns online and outputs the wind turbine frequency command, using the confidence-weighted state vector as the state, the inverter frequency as the action, and the reward consisting of specific energy consumption and emission exceeding penalties as the objective, and enters the next closed loop cycle.

2. The energy efficiency testing method for industrial dust removal based on artificial intelligence as described in claim 1, characterized in that, In step S2, the expected current is determined by dividing the frequency into 2Hz increments when the fan is normal and the filter bag is clean, recording the average current of each increment, and accumulating data from at least three increments to complete the calibration. Piecewise linear interpolation or quadratic polynomial fitting is used to calculate the expected current at any frequency between increments. ； Among them, the two adjacent calibration points are a and b, and the frequency of point a is f a The current is I a The frequency of point b is f b The current is The frequency at any point is f ; The specific steps include: Step S21: Determine the gear position and frequency difference; Step S22: Determine the target frequency offset; Step S23: Determine the current difference; Step S24: Substitute the values ​​into the formula to calculate the ratio and add the intercept.

3. The energy efficiency testing method for industrial dust removal based on artificial intelligence as described in claim 1, characterized in that, In step S3, the pressure difference-concentration consistency index is calculated according to the following formula: ； in, eps It is a constant; The deviation of the pressure-concentration ratio from the historical baseline is calculated using the following formula: ， in, The historical baseline of system resistance corresponding to the measured unit dust concentration. What is the corresponding system resistance for a given unit dust concentration in real time? delta _ R The larger the value, the lower the confidence level of the sensor reading.

4. The energy efficiency testing method for industrial dust removal based on artificial intelligence as described in claim 1, characterized in that, In step S4, the credibility score K is calculated using the following formula: ； in, , , It is the fusion weight of the three types of signals, and + + =1, norm() is a normalization process, clip(...,0,1) limits the result to between 0 and 1 to prevent it from going out of bounds; The closer the value is to 1, the more reliable the dust concentration reading is; The closer it is to 0, the less reliable it is.

5. The energy efficiency testing method for industrial dust removal based on artificial intelligence as described in claim 4, characterized in that, In step S5, the credibility score is... K With a standard preset threshold and low confidence threshold Differentiated and tiered handling based on credibility includes: Step S51, when K≥K th The sensor was deemed reliable and readings were normal, so it was directly... The original value is given to the decision-making process, and the wind turbine continues to adjust intelligently as usual; Step S52, ≤ < Mildly unreliable, downgraded in ranking, no forced correction; Step S53, < =0.3 Severely unreliable, correction, downgrading and alarm are executed simultaneously, the judgment is directly driven by the credibility score to determine the specific fault type; in, It is 0.

3. It is 0.

6.

6. The energy efficiency testing method for industrial dust removal based on artificial intelligence as described in claim 1, characterized in that, In step S7, the confidence-weighted state vector is... i A vector of dimension s i Credibility is alpha i Confidence weight of dust concentration dimension alpha =K, the weighted state is 。 7. The energy efficiency testing method for industrial dust removal based on artificial intelligence as described in claim 1, characterized in that, The reward function r is: ； Where E represents energy consumption and M represents dust removal capacity. λ represents the emission limit, and λ is the penalty coefficient.

8. The energy efficiency testing method for industrial dust removal based on artificial intelligence as described in claim 1, characterized in that, The industrial dust removal energy efficiency testing method also includes a parameter discrimination step: based on the combination of current deviation, pressure difference-concentration consistency deviation and abnormal concentration reading, the specific fault type is determined to be one of the following: low range drift of concentration sensor, interference or disconnection of concentration sensor, abnormality of pressure difference sensor or blockage of sampling tube, or increase of actual load.

9. An industrial dust removal energy efficiency testing system based on artificial intelligence, characterized in that, include: The multi-source data acquisition module is used to synchronously acquire the current, operating frequency, inlet and outlet pressure difference of the dust collector, and measured values ​​of the dust concentration sensor of the main motor of the dust removal fan system. The reliability assessment module is used to calculate the current deviation component based on the physical correspondence between motor current and operating frequency, calculate the pressure difference-concentration consistency deviation based on the coupling relationship between pressure difference and dust concentration, and integrate the current deviation component and the pressure difference-concentration consistency deviation into a reliability score for the dust concentration sensor reading. The concentration prediction module, based on a long short-term memory network, is used to predict future dust concentration trends based on historical concentration sequences. The reinforcement learning decision-making module, based on the SAC algorithm, uses a confidence-weighted state vector as the state, the inverter frequency as the action, and the reward consisting of specific energy consumption and emission exceeding penalties as the objective. It learns online and outputs the wind turbine frequency command. The execution and feedback module is used to send frequency commands to the frequency converter for execution after safety checks, and to store the running data in the database for online model updates; And a module for performing at least one of deviation correction, deweighting or alarm on sensor readings when the confidence score is below a threshold, and feeding the confidence score back to the decision-making process.

10. The industrial dust removal energy-saving efficiency testing system based on artificial intelligence as described in claim 9, characterized in that, The multi-source data acquisition module also includes an assembly rack (2), which is installed on the conveying pipe (1) at the dust collector inlet; The assembly frame (2) is separately equipped with a lens mounting base (3) and a main body mounting base (4), which are coaxially arranged. The assembly frame (2) is provided with a mating groove (21), and the lens mounting base (3) is detachably inserted into the mating groove (21); The assembly frame (2) also has two guide rails (22), which are respectively set on both sides of the mating groove (21), and the main mounting base (4) is slidably set between the two guide rails (22). The main mounting base (4) is suitable for inserting a dust sensor. A limit cover plate (41) is detachably installed on the side of the main mounting base (4) away from the lens mounting base (3). The main mounting base (4) has a light guide plate (42) extending from one side toward the lens mounting base (3), and the light guide plate (42) is inserted into the lens mounting base (3); The inner wall of the lens mounting base (3) away from the main mounting base (4) is provided with several slots (31) along the circumferential direction. The end of the lens mounting base (3) close to the main mounting base (4) is provided with an annular groove (32). The end of the lens mounting base (3) facing the main mounting base (4) is provided with a positioning ring (33). The positioning ring (33) is fitted with the annular groove (32). The positioning ring (33) is provided with several spring pieces (34) along the circumferential direction. The end of the spring piece (34) away from the positioning ring (33) is inserted into the slot (31). The spring (34) is bent toward the center of the lens mounting base (3), and the minimum distance between the spring (34) and the axis is less than the radius of the light guide plate (42); The spring piece (34) is inclinedly arranged on the inner side wall of one end of the slot (31), and a positioning groove (35) is provided on the inner side wall of the spring piece (34). A movable ring (5) is also provided between the lens mounting base (3) and the main body mounting base (4), and the movable ring (5) is sleeved on the outer wall of the light guide plate (42); A number of connecting springs (43) are provided between the movable ring (5) and the main mounting base (4), and the movable ring (5) is suitable for threaded connection to the side wall of the ring groove (32).

Images