Real-time monitoring system for ultrasonic cleaning process of visualizing windows

Abstract

This invention discloses a real-time monitoring system for ultrasonic cleaning processes using a visualization window, relating to the field of charger technology. It includes the following steps: Step S100: Three-dimensional cavitation field modeling: A cavitation bubble point cloud is acquired using a 4D ultrasonic transducer array and a visualization window optical sensing unit, and a three-dimensional cleaning model is constructed by combining multi-physics coupling analysis. By building a dual-dimensional acquisition system of a 4D ultrasonic transducer array and a visualization window optical sensing unit, and combining multi-physics coupling analysis to construct a three-dimensional cavitation field model, a genetic algorithm is used to optimize the transducer phase parameters for different operating frequencies. Combined with 5×5 median filtering and ray tracing algorithms, high-precision denoising, calibration, and registration of multi-source data are achieved. Simultaneously, accurate acquisition of parameters such as flow rate and temperature is achieved through dynamic calibration of standardized sensors.

Description

Technical Field

[0001] This invention relates to the field of ultrasonic cleaning technology, and in particular to a real-time monitoring system for the ultrasonic cleaning process with a visual window. Background Technology

[0002] In modern precision manufacturing processes, ultrasonic cleaning equipment, as a core cleaning device, plays an indispensable role in controlling the surface cleanliness of precision mechanical parts, electronic components, optical devices, and other products. Its working principle is based on the cavitation effect. An ultrasonic transducer converts electrical energy into acoustic energy, generating numerous cavitation bubbles in the cleaning fluid. The energy released when these bubbles burst impacts the surface of the parts, achieving the removal and cleaning of minute stains and impurities. When receiving a working signal from the control unit, the ultrasonic generator adjusts its output power and frequency, and the transducer correspondingly generates cavitation effects of varying intensities, achieving the cleaning objectives for parts of different materials and structures. Ultrasonic cleaning equipment is widely used in many industrial fields such as precision machinery, electronic manufacturing, optical instruments, and aerospace, and is crucial for ensuring the machining accuracy of precision parts and improving product quality.

[0003] Currently, although some ultrasonic cleaning equipment with basic monitoring functions exists on the market, there are still significant technical bottlenecks. Some products lack visual monitoring methods, making it impossible to grasp the distribution of the cavitation field inside the cleaning process in real time. Problems such as cavitation dead zones and excessively high local cavitation intensity are difficult to detect, which can easily lead to uneven cleaning or surface damage to parts. Some can only collect and monitor a single cleaning parameter, lacking the ability to fuse and calibrate multi-source data. The data accuracy is low and the reference value is limited, failing to provide reliable support for the control of the cleaning process. Furthermore, some equipment has a relatively passive approach to anomaly detection and parameter adjustment, relying solely on simple threshold judgments and manual operation. This results in delayed warnings and poor control accuracy, leading not only to low cleaning efficiency and high energy consumption, but also to the risk of parts being scrapped and equipment malfunctioning due to untimely handling of anomalies. Consequently, it is difficult to meet the high-standard and high-consistency cleaning requirements of precision parts. Summary of the Invention

[0004] The purpose of this invention is to provide a real-time monitoring system for ultrasonic cleaning processes with a visual window, which solves the technical problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a real-time monitoring system for ultrasonic cleaning process with a visual window, comprising the following steps: Step S100: 3D cavitation field modeling: A cavitation bubble point cloud is acquired using a 4D ultrasonic transducer array and a visualization window optical sensing unit. A 3D cleaning model is constructed using multiphysics coupling analysis. Sound pressure monitoring points are set at the connection between the transducer array and the cleaning tank. The transducer phase parameters are optimized for different operating frequencies to ensure that the cavitation field uniformity error is ≤8%. Step S200 Multi-source data fusion: Collect data on cleaning fluid flow rate, temperature, conductivity, and cavitation intensity. After reconstruction using a multi-view stereo vision algorithm, register the data using an iterative nearest point algorithm. Denoise and calibrate the image data in the visualization window, and correct boundary refraction deviations using a ray tracing algorithm. Step S300 Abnormal Behavior Detection: Based on the improved anomaly detection model using cost-sensitive filtering and binocular vision calibration parameters, a three-dimensional cleaning trajectory is constructed through triangulation, and the cavitation dead zone and media contamination anomaly are identified by combining the baseline feature space established by K-means clustering. Step S400 Sensor Dynamic Calibration: Arrange sensor arrays around the cleaning tank, pipelines and visualization window, perform spatiotemporal calibration through PTP protocol, and generate a real-time three-dimensional cleaning model containing multi-source data using KD tree algorithm. Step S500 Abnormal fluctuation judgment: The LSTM-GARCH model is used to predict the cleaning parameters. When the prediction confidence is >90%, an early warning is issued. Anomalies are judged in combination with the preset threshold. Step S600 Multimodal Control: When triggered, the ultrasonic generator is linked to adjust the power / frequency and the medium circulation system, and the optimal cleaning strategy is generated through model predictive control algorithm.

[0006] Preferably, in step S100, the piezoelectric acoustic pressure sensor serves as the core component for cavitation field monitoring. It adopts the piezoelectric ceramic stack resonance principle to convert the acoustic pressure signal into an electrical signal output. The sensor has a built-in temperature compensation circuit, which can automatically correct the influence of ambient temperature changes on the measurement results, ensuring that the accuracy is stable at ≤±3% within the 0-30MPa range. During the data acquisition process, the sensor acquires the acoustic pressure data at the connection between the transducer and the tank at a frequency of 10Hz in real time. Through multi-physics coupling analysis software, a coupling model including fluid flow, cavitation effect, and acoustic propagation is constructed. The mesh of dense cavitation areas is dynamically densified using mesh adaptive technology. Phase optimization algorithms are designed for three working frequencies: 40KHz, 80KHz, and 100KHz. Genetic algorithms are introduced for global search during the optimization process. By adjusting the power duty cycle and distribution position of the transducer, the uniformity error of the cavitation field distribution at the corresponding frequency is stably controlled at ≤8%, significantly improving the cleaning uniformity.

[0007] Preferably, in the visualization window image data processing in step S200, the 5×5 median filtering algorithm effectively suppresses bubble interference noise. Its principle is to replace the gray value of each pixel in the image with the median of the gray values ​​of 5×5 pixels in the neighborhood of that point. While preserving the surface details of the part, it removes the interference of cavitation bubbles. In the optical calibration stage, a high-precision standard target plate is used as a reference source. The target plate size accuracy reaches ±0.05mm. By comparing the image acquired in the visualization window with the standard target plate image, a refraction error correction curve is established to make the calibration error ≤±0.5mm. The ray tracing algorithm is based on the Monte Carlo method to simulate the propagation path of light at the glass window-cleaning fluid-air interface. By calculating the reflection, refraction and transmission of light, the point cloud deviation caused by interface refraction is corrected, making the reconstructed three-dimensional cleaning model closer to the actual working conditions.

[0008] Preferably, in step S300, historical data collection covers the operating conditions of ultrasonic cleaning under 3 power levels, 4 common media types, and 5 typical part materials, with a cumulative collection of ≥1000 sets of operating condition data to ensure that the data reflects ≥90% of the actual working scenarios. Through preprocessing of the collected data, key characteristic parameters such as cavitation intensity, flow rate, temperature, and conductivity are extracted. Principal component analysis is used to reduce data dimensionality, retaining more than 95% of the information entropy. Based on the K-means clustering algorithm, the data is clustered into 8-12 operating condition patterns, constructing a benchmark feature space containing the cavitation intensity-cleanliness mapping curve. Simultaneously, in accordance with industry standards for precision parts cleaning, a safe range of cavitation intensity ≥0.2W / cm² is set. 2 The abnormal triggering rules, including the thresholds for flow rate mutation > 0.8 L / min·s, conductivity mutation > 15 μS / cm, and cavitation intensity fluctuation > 15%, were determined through verification of more than 500 sets of experimental data. They can detect cavitation dead zones, media contamination, and abnormal component displacement in a timely and accurate manner.

[0009] Preferably, the sensor calibration in step S400 is crucial. The flow sensor uses an external clamp-on ultrasonic flow meter, calibrated by weighing. Specifically, the cleaning medium is continuously collected for 10 minutes, and the mass of the medium is measured using a high-precision electronic scale with an accuracy of ±0.1g. The actual flow rate is calculated based on the medium density and compared with the sensor measurement value to ensure that the error is ≤±0.5%FS. The temperature sensor is calibrated at a calibration point of 25℃ using a high-precision constant temperature bath with an accuracy of ±0.05℃. By comparing the sensor output value with the measurement value of a standard platinum resistance thermometer, the sensor compensation parameters are adjusted to ensure that the error is ≤±0.2℃. The three-dimensional laser tracker measures the coordinate position of the sensor in space by emitting a laser beam, with a measurement accuracy of ±1mm, effectively ensuring the accuracy of the sensor spatial layout and providing a foundation for building an accurate real-time three-dimensional cleaning model.

[0010] Preferably, in step S500, the LSTM-GARCH model captures the time-series features of the cleaning parameters through a long short-term memory network, and combines this with a generalized autoregressive conditional heteroscedasticity model to handle the volatility and heteroscedasticity of the data, achieving accurate prediction of key parameters such as cavitation intensity and dielectric conductivity. When the predicted value exceeds the normal range and the confidence level is >90%, the system triggers an early warning ≥1 minute in advance, allowing sufficient processing time for maintenance personnel. After multimodal control is triggered, the ultrasonic generator uses a fast-response power adjustment module with a response time <80ms. Combined with a variable frequency water pump, flow rate adjustment is completed within 300ms. By optimizing the transducer phase array and dielectric circulation path, the cavitation intensity recovery rate is ≥0.03W / cm². 2 •s, and the ultrasonic power density remains ≤1.5W / cm² throughout the adjustment process. 2 This effectively avoids damage to parts or inadequate cleaning due to improper cleaning.

[0011] Preferably, the model predictive control algorithm in step S600 aims to maximize cleaning efficiency and minimize part damage. It establishes an optimization problem including a dynamic model of the ultrasonic system, constraints, and an objective function. The objective function J = min(ΔC / ΔP) represents the improvement in part cleanliness and ΔP represents the increment in ultrasonic power consumption. Minimizing this function achieves a balance between cleaning efficiency and energy consumption. Simultaneously, the ultrasonic power density is kept ≤1.5 W / cm². 2 With a pump power of ≤100W as a constraint, the system ensures that the surface of the parts is not damaged and the system energy consumption is reasonable. The MPC algorithm adopts a rolling time-domain optimization strategy, with a control cycle of 1s. In each control cycle, the optimization problem is solved online based on the current system state and the predicted working conditions for the next 5 cycles, generating the optimal power, frequency and flow rate adjustment strategy to achieve coordinated optimization of cleaning performance, energy consumption and part safety.

[0012] Preferably, the visualization window is made of one-piece high borosilicate glass. This glass undergoes chemical strengthening treatment, achieving a surface hardness of Mohs 7 and a light transmittance of ≥95%, ensuring the clarity of the visualization monitoring. The window is sealed to the cleaning tank via a fluororubber O-ring. The sealing surface is precision ground, with a flatness error of ≤0.05mm. It is secured with a metal pressure ring, achieving an IP67 protection rating, effectively preventing cleaning fluid leakage. The temperature resistance range is -10℃ to 90℃, adapting to temperature changes during the cleaning process. The inner side of the window is coated with a nano anti-fog coating with a contact angle of ≥110°, preventing fogging caused by temperature changes during cleaning from affecting observation and ensuring continuous effective visualization monitoring time of ≥95%.

[0013] Preferably, the cleaning media management system adopts a fully sealed circulation structure, uses high-purity deionized water as the base medium, and employs a four-electrode design for the conductivity sensor. It utilizes AC excitation technology to measure the medium conductivity with an accuracy of ≤±1%. When the conductivity exceeds 15μS / cm, the ion exchange resin filtration device is automatically activated. This device employs a two-stage series structure, effectively removing ionic impurities from the medium and restoring its purity to a conductivity of ≤10μS / cm. The level sensor uses the magnetostrictive principle with an accuracy of ≤±1mm, monitoring the cleaning medium level in real time. When the level drops by more than 5%, the replenishment device is automatically triggered to compensate for media loss and ensure the stable operation of the cleaning system.

[0014] Preferably, this real-time monitoring system is specially designed for precision parts cleaning scenarios. All electrical components adopt an IP65 waterproof and moisture-proof design, adaptable to cleaning fluid temperature conditions of 0-90℃. It uses sensing elements and sealing materials with excellent low-temperature performance to ensure normal operation of the system in low-temperature environments. The edge server is equipped with a high-performance processor and an optimized cost-sensitive filtering improved anomaly detection model. This model balances the anomaly missed detection rate and false detection rate by constructing a cost function, with a processing latency of ≤100ms, realizing real-time monitoring and rapid response of the cleaning process. The system supports real-time monitoring of the operating status of single-frequency, dual-frequency, and triple-frequency combined cleaning modes. Cleaning parameters can be set, real-time data and anomaly alarm information can be viewed through the touch screen. It has a cleaning process data traceability function, which can record operating parameters, abnormal events and processing results for ≥90 days. It supports historical data query and export, providing data support for process optimization.

[0015] Compared with related technologies, the real-time monitoring system for ultrasonic cleaning process with a visual window provided by the present invention has the following advantages: 1. This invention provides a real-time monitoring system for ultrasonic cleaning processes with a visualization window. By constructing a dual-dimensional acquisition system of a 4D ultrasonic transducer array and a visualization window optical sensing unit, and combining multi-physics field coupling analysis to construct a three-dimensional cavitation field model, the system utilizes a genetic algorithm to optimize the transducer phase parameters for different operating frequencies. It also employs 5×5 median filtering and ray tracing algorithms to achieve high-precision noise reduction, calibration, and registration of multi-source data. Simultaneously, it achieves accurate acquisition of parameters such as flow rate and temperature through standardized sensor dynamic calibration. This solves the problems of unintuitive monitoring, uneven cavitation field distribution, low accuracy of multi-source data acquisition, and poor fusion effect in traditional ultrasonic cleaning processes, which lead to insufficient uniformity of cleaning of precision parts, incomplete cleaning in some areas, or excessive damage.

[0016] 2. This invention provides a real-time monitoring system for ultrasonic cleaning processes with a visual window. It constructs a benchmark feature space based on over 1000 sets of actual working condition data, and achieves accurate anomaly identification by combining a cost-sensitive filtering-improved anomaly detection model with K-means clustering algorithm. It uses an LSTM-GARCH model to accurately predict cleaning parameters and provide early warnings ≥1 minute in advance. After an anomaly is triggered, a fast-responding ultrasonic generator and variable frequency water pump are linked to complete multimodal control. At the same time, relying on model predictive control algorithm, it generates the optimal control strategy with the goal of maximizing cleaning efficiency and minimizing part damage. This solves the problems of lagging anomaly detection, high false detection and false negative rates, lack of predictability and low accuracy of parameter control in traditional cleaning processes, which lead to high part scrap rates, poor cleaning process stability, and difficulty in balancing efficiency and energy consumption.

[0017] 3. This invention provides a real-time monitoring system for ultrasonic cleaning processes with a visual window. By designing a highly adaptable hardware structure, it adopts a chemically reinforced high borosilicate glass visual window combined with a fluororubber sealing structure to achieve IP67 protection. All electrical components adopt an IP65 waterproof and moisture-proof design to adapt to complex temperature change conditions. A fully sealed cleaning medium circulation management system is built to achieve intelligent control of medium purity and liquid level. At the same time, it integrates data traceability, multi-mode monitoring and process parameter visualization setting functions. It solves the problems of poor adaptability of traditional cleaning systems to complex working conditions of precision parts cleaning, easy contamination of media leading to frequent equipment failures, lack of full-process data support to achieve process optimization, high manual maintenance costs and low production management efficiency. Attached Figure Description

[0018] Figure 1 This is a flowchart of the present invention; Figure 2 This is an extended flowchart of the three-dimensional cavitation field modeling of the present invention; Figure 3 This is an extended flowchart of the multi-source data fusion method of the present invention; Figure 4 This is an extended flowchart of the abnormal behavior detection method of the present invention; Figure 5 This is an extended flowchart of the sensor dynamic calibration of the present invention; Figure 6 This is an extended flowchart of the abnormal fluctuation judgment method of the present invention; Figure 7 This is an extended flowchart of the multimodal control method of the present invention. Detailed Implementation

[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0020] Example 1: Please see Figures 1-7 This invention provides a technical solution: a real-time monitoring system for ultrasonic cleaning processes with a visual window, comprising the following steps: Step S100: 3D cavitation field modeling: A cavitation bubble point cloud is acquired using a 4D ultrasonic transducer array and a visualization window optical sensing unit. A 3D cleaning model is constructed using multiphysics coupling analysis. Sound pressure monitoring points are set at the connection between the transducer array and the cleaning tank. The transducer phase parameters are optimized for different operating frequencies to ensure that the cavitation field uniformity error is ≤8%. In step S100, the piezoelectric acoustic pressure sensor serves as the core component for cavitation field monitoring. It adopts the piezoelectric ceramic stack resonance principle to convert the acoustic pressure signal into an electrical signal output. The sensor has a built-in temperature compensation circuit, which can automatically correct the influence of ambient temperature changes on the measurement results, ensuring that the accuracy is stable at ≤±3% within the 0-30MPa range. During the data acquisition process, the sensor acquires the acoustic pressure data at the connection between the transducer and the tank at a frequency of 10Hz in real time. Through multiphysics coupling analysis software, a coupling model including fluid flow, cavitation effect and acoustic propagation is constructed. The mesh of dense cavitation area is dynamically densified using mesh adaptive technology. Phase optimization algorithms are designed for three working frequencies of 40KHz, 80KHz and 100KHz respectively. Genetic algorithm is introduced for global search during the optimization process. By adjusting the power duty cycle and distribution position of the transducer, the uniformity error of the cavitation field distribution at the corresponding frequency is stably controlled at ≤8%, which significantly improves the cleaning uniformity. In this implementation scheme, the 4D ultrasonic transducer array adopts a 16-channel distributed layout, evenly installed at the bottom and sides of the cleaning tank. Combined with a visualization window optical sensing unit, it achieves dual-dimensional data acquisition of cavitation bubbles using both acoustic and optical methods. Acoustic data captures the bubble vibration intensity, while optical data records the spatial distribution of the bubbles. This combination improves the 3D reconstruction accuracy of the cavitation field point cloud by more than 30%. For the three core frequencies of 40kHz, 80kHz, and 100kHz, the phase optimization algorithm iterates more than 200 times using a genetic algorithm, ultimately stabilizing the cavitation field uniformity error at each frequency to 6%-8%. Compared to the traditional fixed-phase scheme, this improves cleaning uniformity by 40%, effectively avoiding the problem of incomplete cleaning or excessive damage to parts in certain areas.

[0021] Step S200 Multi-source data fusion: Collect data on cleaning fluid flow rate, temperature, conductivity, and cavitation intensity. After reconstruction using a multi-view stereo vision algorithm, register the data using an iterative nearest point algorithm. Denoise and calibrate the image data in the visualization window, and correct boundary refraction deviations using a ray tracing algorithm. In the S200 step, the image data processing of the visualization window effectively suppresses bubble interference noise using a 5×5 median filtering algorithm. The principle is to replace the gray value of each pixel in the image with the median of the gray values ​​of 5×5 pixels in the neighborhood of that pixel. This removes cavitation bubble occlusion interference while preserving the surface details of the part. In the optical calibration step, a high-precision standard target plate is used as a reference source. The target plate size accuracy reaches ±0.05mm. By comparing the image acquired through the visualization window with the standard target plate image, a refraction error correction curve is established to ensure that the calibration error is ≤±0.5mm. The ray tracing algorithm is based on the Monte Carlo method to simulate the propagation path of light at the glass window-cleaning fluid-air interface. By calculating the reflection, refraction, and transmission of light, the point cloud deviation caused by interface refraction is corrected, making the reconstructed three-dimensional cleaning model closer to the actual working conditions. In this implementation scheme, multi-source data fusion adopts a "layered processing + spatiotemporal registration" strategy: First, the visualization image is preprocessed using a 5×5 median filter, which improves noise suppression by 25% in scenes with dense bubbles compared to the traditional 3×3 filter, while preserving details of scratches and stains on the part surface at the 0.1mm level. In the optical calibration stage, the standard target plate adopts a cross-scale design, and a refraction error matrix is ​​constructed by acquiring images from 5 different angles. After correction using a ray tracing algorithm, the point cloud deviation is reduced from ±1.2mm to within ±0.5mm. After multi-source data is registered using an iterative nearest-point algorithm, the time synchronization error is ≤10ms and the spatial registration error is ≤0.3mm, providing high-precision data support for the real-time updating of the subsequent 3D cleaning model.

[0022] Step S300 Abnormal Behavior Detection: Based on the improved anomaly detection model using cost-sensitive filtering and binocular vision calibration parameters, a three-dimensional cleaning trajectory is constructed through triangulation, and the cavitation dead zone and media contamination anomaly are identified by combining the baseline feature space established by K-means clustering. In step S300, historical data collection covers the operating conditions of ultrasonic cleaning under 3 power levels, 4 common media types, and 5 typical part materials, accumulating ≥1000 sets of operating condition data to ensure that the data reflects ≥90% of actual working scenarios. Through data preprocessing, key characteristic parameters such as cavitation intensity, flow rate, temperature, and conductivity are extracted. Principal component analysis is used to reduce data dimensionality, retaining over 95% of the information entropy. Based on the K-means clustering algorithm, the data is clustered into 8-12 operating condition patterns, constructing a benchmark feature space including the cavitation intensity-cleanliness mapping curve. Simultaneously, in accordance with industry standards for precision parts cleaning, a safe range of cavitation intensity ≥0.2W / cm² is set. 2 The thresholds for abnormal triggering rules, such as flow rate mutation > 0.8 L / min·s, conductivity mutation > 15 μS / cm, and cavitation intensity fluctuation > 15%, were determined through verification of more than 500 sets of experimental data. They can detect cavitation dead zones, media contamination, and abnormal component displacement in a timely and accurate manner. In this implementation plan, historical operating condition data collection covers three power levels (20W / 40W / 60W), four commonly used media (deionized water / anhydrous ethanol / alkaline cleaning solution / neutral cleaning solution), and five typical component materials (stainless steel / aluminum alloy / ceramic / glass / plastic). Through combined tests at different temperatures and cleaning times, a total of 1200 sets of valid operating condition data were acquired, ensuring that the baseline feature space can cover 92% of actual application scenarios. In the cost-sensitive filtering-improved anomaly detection model, the cost function is set as J = 1.5 × false negative rate + 0.8 × false positive rate. Validated by 500 sets of experimental data, the anomaly detection accuracy is ≥96%, the cavitation dead zone identification delay is ≤0.5s, and the media contamination early warning response time is ≤1s, effectively reducing the component scrap rate caused by anomalies.

[0023] Step S400 Sensor Dynamic Calibration: Arrange sensor arrays around the cleaning tank, pipelines and visualization window, perform spatiotemporal calibration through PTP protocol, and generate a real-time three-dimensional cleaning model containing multi-source data using KD tree algorithm. Sensor calibration in the S400 process is crucial. The flow sensor uses an external clamp-on ultrasonic flow meter, calibrated by weighing. Specifically, the cleaning medium is continuously collected for 10 minutes, and the mass of the medium is measured using a high-precision electronic scale with an accuracy of ±0.1g. The actual flow rate is calculated based on the medium density and compared with the sensor measurement value to ensure the error is ≤±0.5%FS. The temperature sensor is calibrated at a calibration point of 25℃ using a high-precision constant temperature bath with an accuracy of ±0.05℃. By comparing the sensor output value with the measurement value of a standard platinum resistance thermometer, the sensor compensation parameters are adjusted to ensure the error is ≤±0.2℃. The three-dimensional laser tracker measures the coordinate position of the sensor in space by emitting a laser beam, with a measurement accuracy of ±1mm, effectively ensuring the accuracy of the sensor spatial layout and providing a foundation for building an accurate real-time three-dimensional cleaning model. In this implementation scheme, the sensor array includes six piezoelectric acoustic pressure sensors, two clamp-on ultrasonic flow meters, four temperature sensors, and one conductivity sensor, evenly distributed in key areas of the cleaning tank. Spatiotemporal synchronization of all sensors is achieved via the PTP protocol, with a time calibration error ≤5ms. Spatial calibration is performed using a three-dimensional laser tracker, with the deviation between the actual sensor installation position and the designed position ≤0.8mm. After calibration, the flow measurement error stabilizes within ±0.3%FS-±0.5%FS, and the temperature measurement error ≤±0.15℃, ensuring the parameter accuracy of the real-time three-dimensional cleaning model and providing a reliable data foundation for anomaly detection and control decisions.

[0024] Step S500 Abnormal fluctuation judgment: The LSTM-GARCH model is used to predict the cleaning parameters. When the prediction confidence is >90%, an early warning is issued. Anomalies are judged in combination with the preset threshold. In the S500 step, the LSTM-GARCH model captures the time-series characteristics of the cleaning parameters through a long short-term memory network. Combined with a generalized autoregressive conditional heteroscedasticity model to handle data volatility and heteroscedasticity, it achieves accurate prediction of key parameters such as cavitation intensity and dielectric conductivity. When the predicted value exceeds the normal range with a confidence level >90%, the system triggers an early warning at least 1 minute in advance, providing sufficient processing time for maintenance personnel. After multimodal control is triggered, the ultrasonic generator uses a fast-response power adjustment module with a response time <80ms. Combined with a variable frequency water pump, flow rate adjustment is completed within 300ms. By optimizing the transducer phase array and dielectric circulation path, the cavitation intensity recovery rate is ≥0.03W / cm². 2 •s, and the ultrasonic power density remains ≤1.5W / cm² throughout the adjustment process. 2 This effectively avoids damage to parts or inadequate cleaning due to abnormal cleaning. In this implementation scheme, the training dataset for the LSTM-GARCH model is divided into training and testing sets in a 7:3 ratio. During training, the adaptive learning rate is adjusted to ensure that the model's prediction error is ≤3%. When the system detects cavitation intensity below 0.2 W / cm²... 2 When the conductivity suddenly increases by more than 15 μS / cm or the flow rate suddenly changes by more than 0.8 L / min·s, an early warning is immediately triggered if the confidence level of the LSTM-GARCH model predicting the abnormality to persist for the next 5 seconds is greater than 90%. During the multimodal control response, the power regulation module of the ultrasonic generator adopts an IGBT inverter circuit design, with a response time stable at 60ms-80ms. The variable frequency water pump achieves rapid flow regulation through PWM speed control, completing flow rate switching from 0.5L / min to 2L / min within 300ms, and the cavitation intensity recovery rate reaches 0.03 W / cm². 2 ・s-0.05W / cm 2 •s, and the ultrasonic power density is controlled at 1.2W / cm² throughout the entire process. 2 -1.5W / cm 2 This ensures effective cleaning while preventing damage to parts.

[0025] Step S600 Multimodal Control: When triggered, the ultrasonic generator is linked to adjust the power / frequency and the medium circulation system, and the optimal cleaning strategy is generated through model predictive control algorithm; The model predictive control algorithm in step S600 aims to maximize cleaning efficiency and minimize part damage. It establishes an optimization problem including a dynamic model of the ultrasonic system, constraints, and an objective function: J = min(ΔC / ΔP), where ΔC represents the improvement in part cleanliness and ΔP represents the increment in ultrasonic power consumption. Minimizing this function achieves a balance between cleaning efficiency and energy consumption, while simultaneously ensuring the ultrasonic power density is ≤1.5 W / cm². 2 With a pump power of ≤100W as a constraint, the surface of the parts is not damaged and the system energy consumption is reasonable. The MPC algorithm adopts a rolling time-domain optimization strategy, with a control cycle of 1s. In each control cycle, the optimization problem is solved online based on the current system state and the predicted working conditions of the next 5 cycles, generating the optimal power, frequency and flow rate adjustment strategy to achieve coordinated optimization of cleaning performance, energy consumption and part safety. In this implementation scheme, in the objective function J=min(ΔC / ΔP) of the model predictive control algorithm, ΔC is obtained through the surface cleanliness detection of the part, and ΔP is acquired through the power monitoring module of the ultrasonic generator. The constraint condition is that the ultrasonic power density is ≤1.5W / cm³. 2Referring to the damage threshold requirements of GB / T30038-2013 "General Rules for Cleaning Precision Mechanical Parts", the pump power ≤100W is achieved by optimizing the pump impeller structure and pipeline resistance. The control cycle of the MPC algorithm is set to 1s, predicting the operating condition changes in the next 5s within each cycle, and solving the optimal control parameters through a quadratic programming algorithm. Compared with traditional PID control, the cleaning efficiency is improved by 20%, energy consumption is reduced by 15%, and the consistency error of part cleanliness is ≤3%, achieving synergistic optimization of cleaning performance, energy consumption, and part safety.

[0026] The visualization window is made of one-piece high borosilicate glass, which has undergone chemical strengthening treatment, achieving a surface hardness of 7 on the Mohs scale and a light transmittance of ≥95%, ensuring the clarity of the visualization monitoring. The window is sealed to the cleaning tank via a fluororubber O-ring, and the sealing surface is precision ground to achieve a flatness error of ≤0.05mm. It is secured with a metal pressure ring, achieving an IP67 protection rating to effectively prevent cleaning fluid leakage. The temperature resistance range is -10℃ to 90℃, adapting to temperature changes during the cleaning process. The inner side of the window is coated with a nano anti-fog coating with a contact angle of ≥110°, preventing fogging caused by temperature changes during cleaning from affecting observation and ensuring continuous effective visualization monitoring time of ≥95%. The cleaning media management system adopts a fully sealed circulation structure, using high-purity deionized water as the base medium. The conductivity sensor uses a four-electrode design and AC excitation technology to measure the medium conductivity with an accuracy of ≤±1%. When the conductivity is >15μS / cm, the ion exchange resin filtration device is automatically activated. This device uses a two-stage series structure, which can effectively remove ionic impurities in the medium and restore the medium purity to a conductivity of ≤10μS / cm. The liquid level sensor uses the magnetostrictive principle with an accuracy of ≤±1mm to monitor the cleaning medium level in real time. When the liquid level drops by more than 5%, the liquid replenishment device is automatically triggered to compensate for media loss and ensure the stable operation of the cleaning system. This real-time monitoring system is specially designed for precision parts cleaning scenarios. All electrical components are IP65 waterproof and moisture-proof, adaptable to cleaning fluid temperatures ranging from 0-90℃. It employs low-temperature performance sensors and sealing materials to ensure normal operation in low-temperature environments. The edge server is equipped with a high-performance processor and an optimized cost-sensitive filtering improved anomaly detection model. This model balances the anomaly miss rate and false detection rate by constructing a cost function, with a processing latency of ≤100ms, enabling real-time monitoring and rapid response of the cleaning process. The system supports real-time monitoring of the operating status of single-frequency, dual-frequency, and triple-frequency combined cleaning modes. Cleaning parameters can be set, real-time data and anomaly alarm information can be viewed via a touch screen. It has a cleaning process data traceability function, which can record operating parameters, abnormal events and processing results for ≥90 days, and supports historical data query and export, providing data support for process optimization.

[0027] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium is implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0028] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention should all be covered within the scope of the claims of the present invention.

Claims

1. A real-time monitoring system for ultrasonic cleaning process with a visual window, characterized in that, Includes the following steps: Step S100: 3D cavitation field modeling: A cavitation bubble point cloud is acquired using a 4D ultrasonic transducer array and a visualization window optical sensing unit. A 3D cleaning model is constructed using multiphysics coupling analysis. Sound pressure monitoring points are set at the connection between the transducer array and the cleaning tank. The transducer phase parameters are optimized for different operating frequencies to ensure that the cavitation field uniformity error is ≤8%. Step S200 Multi-source data fusion: Collect data on cleaning fluid flow rate, temperature, conductivity and cavitation intensity, reconstruct it using a multi-view stereo vision algorithm, and then register it using an iterative nearest point algorithm; The visualization window image data is denoised and calibrated, and the boundary refraction deviation is corrected by combining ray tracing algorithm; Step S300 Abnormal Behavior Detection: Based on the improved anomaly detection model using cost-sensitive filtering and binocular vision calibration parameters, a three-dimensional cleaning trajectory is constructed through triangulation, and the cavitation dead zone and media contamination anomaly are identified by combining the baseline feature space established by K-means clustering. Step S400 Sensor Dynamic Calibration: Arrange sensor arrays around the cleaning tank, pipelines and visualization window, perform spatiotemporal calibration through PTP protocol, and generate a real-time three-dimensional cleaning model containing multi-source data using KD tree algorithm. Step S500 Abnormal fluctuation judgment: The LSTM-GARCH model is used to predict the cleaning parameters. When the prediction confidence is >90%, an early warning is issued. Anomalies are judged in combination with the preset threshold. Step S600 Multimodal Control: When triggered, the ultrasonic generator is linked to adjust the power / frequency and the medium circulation system, and the optimal cleaning strategy is generated through model predictive control algorithm.

2. The real-time monitoring system for ultrasonic cleaning process with a visual window according to claim 1, characterized in that: In step S100, the piezoelectric acoustic pressure sensor serves as the core component for cavitation field monitoring. Utilizing the piezoelectric ceramic resonant principle, it converts acoustic pressure signals into electrical signals. The sensor incorporates a temperature compensation circuit, automatically correcting for the impact of ambient temperature changes on measurement results, ensuring a stable accuracy of ≤±3% within the 0-30MPa range. During data acquisition, the sensor acquires real-time acoustic pressure data at the transducer-tank connection point at a frequency of 10Hz. Multiphysics coupling analysis software is used to construct a coupled model incorporating fluid flow, cavitation effects, and acoustic propagation. Adaptive mesh technology is employed to dynamically densify the mesh in dense cavitation areas. Phase optimization algorithms are designed for three operating frequencies: 40kHz, 80kHz, and 100kHz. A genetic algorithm is introduced for global search during optimization. By adjusting the transducer power duty cycle and distribution position, the uniformity error of the cavitation field distribution at the corresponding frequencies is stably controlled within ≤8%, significantly improving cleaning uniformity.

3. The real-time monitoring system for ultrasonic cleaning process with a visual window according to claim 1, characterized in that: In the S200 step, the image data processing of the visualization window effectively suppresses bubble interference noise using a 5×5 median filtering algorithm. This algorithm replaces the gray value of each pixel in the image with the median of the gray values ​​of 5×5 pixels in the neighborhood of that pixel, preserving the surface details of the part while removing interference from cavitation bubbles. In the optical calibration stage, a high-precision standard target plate is used as a reference source, with a target plate size accuracy of ±0.05mm. By comparing the image acquired through the visualization window with the standard target plate image, a refraction error correction curve is established, ensuring the calibration error is ≤±0.5mm. The ray tracing algorithm, based on the Monte Carlo method, simulates the propagation path of light at the glass window-cleaning fluid-air interface. By calculating the reflection, refraction, and transmission of light, it corrects the point cloud deviation caused by interface refraction, making the reconstructed 3D cleaning model closer to actual working conditions.

4. The real-time monitoring system for ultrasonic cleaning process with a visual window according to claim 1, characterized in that: In step S300, historical data collection covers the operating conditions of ultrasonic cleaning under 3 power levels, 4 common media types, and 5 typical part materials, accumulating ≥1000 sets of operating condition data to ensure that the data reflects ≥90% of actual working scenarios. Through data preprocessing, key characteristic parameters such as cavitation intensity, flow rate, temperature, and conductivity are extracted. Principal component analysis is used to reduce data dimensionality, retaining over 95% of the information entropy. Based on the K-means clustering algorithm, the data is clustered into 8-12 operating condition patterns, constructing a benchmark feature space including the cavitation intensity-cleanliness mapping curve. Simultaneously, in accordance with industry standards for precision parts cleaning, a safe range of cavitation intensity ≥0.2W / cm² is set. 2 The abnormal triggering rules, including the thresholds for flow rate mutation > 0.8 L / min·s, conductivity mutation > 15 μS / cm, and cavitation intensity fluctuation > 15%, were determined through verification of more than 500 sets of experimental data. They can detect cavitation dead zones, media contamination, and abnormal component displacement in a timely and accurate manner.

5. The real-time monitoring system for ultrasonic cleaning process with a visual window according to claim 1, characterized in that: Sensor calibration in the S400 process is crucial. The flow sensor uses an external clamp-on ultrasonic flow meter, calibrated by weighing. Specifically, the cleaning medium is continuously collected for 10 minutes, and the mass of the medium is measured using a high-precision electronic scale with an accuracy of ±0.1g. The actual flow rate is calculated based on the medium density and compared with the sensor measurement to ensure an error ≤±0.5%FS. The temperature sensor is calibrated at a calibration point of 25℃ using a high-precision constant temperature bath with an accuracy of ±0.05℃. By comparing the sensor output value with the measurement value of a standard platinum resistance thermometer, the sensor compensation parameters are adjusted to ensure an error ≤±0.2℃. The 3D laser tracker measures the coordinate position of the sensor in space by emitting a laser beam, with a measurement accuracy of ±1mm, effectively ensuring the accuracy of the sensor's spatial layout and providing a foundation for building an accurate real-time 3D cleaning model.

6. The real-time monitoring system for ultrasonic cleaning process with a visual window according to claim 1, characterized in that: In the S500 step, the LSTM-GARCH model captures the time-series characteristics of the cleaning parameters through a long short-term memory network. Combined with a generalized autoregressive conditional heteroscedasticity model to handle data volatility and heteroscedasticity, it achieves accurate prediction of key parameters such as cavitation intensity and dielectric conductivity. When the predicted value exceeds the normal range with a confidence level >90%, the system triggers an early warning at least 1 minute in advance, providing sufficient processing time for maintenance personnel. After multimodal control is triggered, the ultrasonic generator uses a fast-response power adjustment module with a response time <80ms. Combined with a variable frequency water pump, flow rate adjustment is completed within 300ms. By optimizing the transducer phase array and dielectric circulation path, the cavitation intensity recovery rate is ≥0.03W / cm². 2 •s, and the ultrasonic power density remains ≤1.5W / cm² throughout the adjustment process. 2 This effectively avoids damage to parts or inadequate cleaning due to abnormal cleaning.

7. The real-time monitoring system for ultrasonic cleaning process with a visual window according to claim 1, characterized in that: The model predictive control algorithm in step S600 aims to maximize cleaning efficiency and minimize part damage. It establishes an optimization problem including a dynamic model of the ultrasonic system, constraints, and an objective function: J = min(ΔC / ΔP), where ΔC represents the improvement in part cleanliness and ΔP represents the increment in ultrasonic power consumption. Minimizing this function achieves a balance between cleaning efficiency and energy consumption, while simultaneously ensuring the ultrasonic power density is ≤1.5 W / cm². 2 With a pump power of ≤100W as a constraint, the system ensures that the surface of the parts is not damaged and the system energy consumption is reasonable. The MPC algorithm adopts a rolling time-domain optimization strategy, with a control cycle of 1s. In each control cycle, the optimization problem is solved online based on the current system state and the predicted working conditions for the next 5 cycles, generating the optimal power, frequency and flow rate adjustment strategy to achieve coordinated optimization of cleaning performance, energy consumption and part safety.

8. The real-time monitoring system for ultrasonic cleaning process with a visual window according to claim 1, characterized in that: The visualization window is made of high borosilicate glass in one piece. This glass has undergone chemical strengthening treatment, achieving a surface hardness of 7 on the Mohs scale and a light transmittance of ≥95%, ensuring the clarity of the visualization monitoring. The window is sealed to the cleaning tank via a fluororubber O-ring. The sealing surface is processed with precision grinding, achieving a flatness error of ≤0.05mm. It is then secured with a metal pressure ring, providing an IP67 protection rating to effectively prevent cleaning fluid leakage. The temperature resistance range is -10℃ to 90℃, adapting to temperature changes during the cleaning process. The inner side of the window is coated with a nano anti-fog coating with a contact angle of ≥110°, preventing fogging caused by temperature changes during the cleaning process from affecting observation and ensuring that the continuous effective time of visualization monitoring is ≥95%.

9. The real-time monitoring system for ultrasonic cleaning process with a visual window according to claim 1, characterized in that: The cleaning media management system adopts a fully sealed circulation structure, using high-purity deionized water as the base medium. The conductivity sensor uses a four-electrode design and AC excitation technology to measure the medium conductivity with an accuracy of ≤±1%. When the conductivity is >15μS / cm, the ion exchange resin filtration device is automatically activated. This device uses a two-stage series structure, which can effectively remove ionic impurities in the medium and restore the medium purity to a conductivity of ≤10μS / cm. The liquid level sensor uses the magnetostrictive principle with an accuracy of ≤±1mm to monitor the cleaning medium level in real time. When the liquid level drops by more than 5%, the liquid replenishment device is automatically triggered to compensate for media loss and ensure the stable operation of the cleaning system.

10. A real-time monitoring system for ultrasonic cleaning process with a visual window according to claim 1, characterized in that: This real-time monitoring system is specially designed for precision parts cleaning scenarios. All electrical components are IP65 waterproof and moisture-proof, adaptable to cleaning fluid temperatures ranging from 0-90℃. It employs low-temperature performance sensors and sealing materials to ensure normal operation in low-temperature environments. The edge server is equipped with a high-performance processor and an optimized cost-sensitive filtering improved anomaly detection model. This model balances the anomaly miss rate and false detection rate by constructing a cost function, with a processing latency of ≤100ms, enabling real-time monitoring and rapid response of the cleaning process. The system supports real-time monitoring of the operating status of single-frequency, dual-frequency, and triple-frequency combined cleaning modes. Cleaning parameters can be set, real-time data and anomaly alarm information can be viewed via a touch screen. It has a cleaning process data traceability function, which can record operating parameters, abnormal events and processing results for ≥90 days, and supports historical data query and export, providing data support for process optimization.

Images