Micro-plastic high-flux mass spectrum detection device and method for synchronously testing multi-dimensional characteristics

Abstract

The invention provides a micro-plastic high-flux mass spectrum detection device and method for multi-dimensional characteristic synchronous testing, and relates to the technical field of micro-plastic pollution monitoring, the micro-plastic high-flux mass spectrum detection device comprises a rapid image acquisition and analysis module, a gas circuit transmission module, an electromagnetic thermal cracking module, a filtration and purification module, a mass spectrum detection module and a data analysis module, the gas circuit transmission module, the electromagnetic thermal cracking module, the filtration and purification module and the mass spectrometry detection module are sequentially communicated through quartz pipelines, the rapid image acquisition and analysis module is arranged right above the electromagnetic thermal cracking module and is separated by a visible window, a window shielding gas purging opening is formed below the visible window, and the electromagnetic thermal cracking module is connected with a shunt sucking pump. According to the invention, four-dimensional information of particle size, quantity, variety and quality of the micro-plastic in the sample is synchronously obtained, the accuracy of multi-dimensional information acquisition of the micro-plastic can be improved, conversion between quality and quantity of the micro-plastic is completed, and the problems that a window of a rapid image acquisition and analysis module is polluted and a microscope lens is not resistant to high temperature are solved through gas purging.

Description

Technical Field

[0001] This application relates to the technical field of microplastic pollution monitoring, and more specifically, to a high-throughput mass spectrometry detection device and method for microplastics with synchronous testing of multi-dimensional characteristics. Background Art

[0002] In the global ecological environment issues, microplastic pollution has become a focus. Plastic debris, fibers or particles with a size ≤ 5 mm are collectively referred to as microplastics. Microplastics are also known as "PM2.5 in the ocean" and are listed as major global environmental problems together with global climate change, ozone depletion and ocean acidification. They cause serious harm to the marine environment and ecology. Microplastics suspended in seawater will affect the photosynthesis of marine algae; they are easily ingested by some marine organisms and are difficult to digest in the gastrointestinal tract, causing malnutrition or even death of marine organisms. Some studies have found that microplastics can also affect the spawning and reproductive ability of some marine organisms such as fish; in addition, microplastics themselves contain some toxic plastic additives, and their surfaces are also prone to enrichment of organic pollutants and heavy metals, etc., which will have a toxic effect on surrounding organisms. Characteristics such as the type, number concentration, particle size and mass concentration of microplastics jointly determine the environmental geochemical process of microplastics. Therefore, the identification of microplastics is the key to deeply exploring microplastic pollution and improving microplastic pollution control.

[0003] As a unique type of polymer particles, compared with traditional pollutants, the characterization of microplastics is difficult to be accurately completed by a single physical or chemical analysis method. Instead, it requires a comprehensive assessment of its physical morphology (such as size, shape) and polymerization characteristics (such as chemical composition, surface properties). At present, the detection of microplastics mainly uses visual method, microscopy, spectroscopy and thermal analysis mass spectrometry to characterize important characteristics such as the particle size, quantity, type and mass of microplastics. Although the above methods have their own advantages, they all have limitations. The visual method is labor-intensive and difficult to quantify. The spectroscopy is affected by sample characteristics and has a long analysis time. To further identify more dimensional information of microplastics, microspectroscopy combines microscopy imaging technology with infrared spectroscopy or Raman spectroscopy to identify three-dimensional information of particle size, type and quantity. However, there is a lack of acquisition of microplastic mass information, and it may encounter challenges of overlapping test information and decreased accuracy. Thermal analysis mass spectrometry combines pyrolysis technology and mass spectrometry technology to analyze two-dimensional information of the type and mass of microplastics. However, due to its destructive analysis characteristics, the particle size and quantity information of microplastics cannot be obtained. If multiple detection methods are used jointly to obtain four-dimensional information of microplastic quantity, type, particle size and mass, it will lead to poor comparability of experimental data between different sources and reduce the credibility of the final evaluation results. And the excessively long test time challenges high-throughput testing, seriously restricting the efficiency of sample testing and making it difficult to conduct environmental effect assessment based on a large sample size. To sum up, with the current detection technology level, it is difficult to synchronously and quickly capture multi-dimensional characteristics such as the quantity, type, particle size and mass of microplastics, especially there are great difficulties in breaking through the synchronous acquisition of quantity and mass concentration. This technical bottleneck has led to little substantial progress in the migration and transformation of microplastics and their safety assessment in the fields of environment, food, medicine, health, etc.

[0004] Therefore, there is an urgent need to develop a method that can quickly and simultaneously test four-dimensional information of microplastic particle size, quantity, type and mass to solve the problems that it is difficult to synchronously identify multi-dimensional characteristics of microplastics and high-throughput cannot be achieved. Summary of the Invention

[0005] To solve the above problems, the technical solution adopted in this application is a high-throughput mass spectrometry detection device for synchronous testing of multi-dimensional characteristics of microplastics, including: a fast image acquisition and analysis module, a gas path transmission module, an electromagnetic pyrolysis module, a filtration and purification module, a mass spectrometry detection module and a data analysis module. The gas path transmission module, the electromagnetic pyrolysis module, the filtration and purification module and the mass spectrometry detection module are sequentially connected through a quartz pipeline. The fast image acquisition and analysis module is arranged directly above the electromagnetic pyrolysis module and separated by a visual window. A window shielding gas purge port is arranged below the visual window. The electromagnetic pyrolysis module is connected with a shunt air extraction pump.

[0006] Optionally, the rapid image acquisition and analysis module includes a microscope, a camera, and an image processing device; the gas path transmission module includes a carrier gas source and a control valve; the electromagnetic thermal cracking module includes a thermal cracking quartz chamber, an electromagnetic induction coil, a quartz stage, and a stage support; the filtration and purification module includes quartz fibers and a metal filter; the mass spectrometry detection module is equipped with a mass spectrometry detection and analysis device with the ability to detect volatile gases.

[0007] This application also provides a high-throughput mass spectrometry detection method for microplastics with multi-dimensional feature synchronous testing. Using the high-throughput microplastic mass spectrometry detection device for multi-dimensional feature synchronous detection as described in any of the foregoing, it includes the following steps: S1: Place the microplastic sample in the electromagnetic thermal cracking module, and collect the sample image through the rapid image acquisition and analysis module; S2: Provide a vacuum environment for the electromagnetic thermal cracking module through a shunt air pump, and then crack the sample through the electromagnetic thermal cracking module. The gas path transmission module provides carrier gas to the electromagnetic thermal cracking module. After the thermal cracking gas products of the microplastics are purified by the filtration and purification module, they are transported into the mass spectrometry detection module for analysis, and shielding gas is blown through the visual window shielding gas purge port; S3: The rapid image acquisition and analysis module provides the surface microplastic particle size distribution information for the data analysis module, the mass spectrometry detection module provides the overall microplastic mass and type information for the data analysis module, and the data analysis module analyzes through the microplastic multi-parameter machine learning model.

[0008] Optionally, the construction of the microplastic multi-parameter machine learning model includes: S4-1: Obtain actual test data, optimize the data through the microplastic shape density calibration program to generate reliable output data. The reliable output data forms a true value simulation training data set for machine learning through feature vector splicing, and the true value simulation training data set is divided into a training set and a test set; S4-2: Establish a neural network model including an input layer, a hidden layer, a physical constraint fusion layer, and an output layer; S4-3: Train the neural network model through the training set through the initialization model, model instantiation, definition of the loss function, and training loop and optimization process; S4-4: Verify the neural network model through the test set and evaluate the model performance.

[0009] Optionally, obtaining actual test data includes: using microplastic standards and actual samples, and obtaining the mass, type quantity, and particle size information of microplastics using the high-throughput microplastic mass spectrometry detection device for multi-dimensional feature synchronous detection.

[0010] Optionally, the data optimization to generate reliable output data includes calculating the image volume and density distribution based on the particle size distribution histogram and the empirical shape factor respectively, and the mass spectrometry volume based on the mass and density of microplastics, performing volume constraint, and directly using the data that meets the constraint conditions as the reliable output data, which enters the machine learning prediction channel to obtain the output value of the final total number of particles; the data that does not meet the constraint conditions enters the volume constraint optimization channel, and after the joint optimization of the shape factor and density, reliable output data is formed.

[0011] Optionally, the calculation of the mass spectrometry volume is carried out according to the following formula: ; In the formula, V M is the volume of K types of microplastics obtained by mass spectrometry test calculation, M k is the mass of K types of microplastics obtained by mass spectrometry test; ρ k is the density of K types of microplastics; The calculation of the image volume is carried out according to the following formula: ; In the formula, V E is the volume of K types of microplastics expected to be obtained by microplastic recognition in the image; ρ k is the density of K types of microplastics, which is a fixed value; n i is the proportion of the number of particles in each particle size interval; d i is the diameter of the particle size interval; h is the shape factor; P ki is the proportion of K types of plastics in the i particle size interval; 1 / ρ k is the volume conversion factor, which converts the proportion of quantity into the proportion of volume.

[0012] Optionally, the physical constraint fusion includes mapping the total mass of microplastics obtained by mass spectrometry test to the same dimension as the hidden layer features through a fully connected layer, and multiplying the mapped total mass features and the hidden layer features element by element to achieve physical constraint fusion; the fused features are passed to the output layer to predict the number of microplastics.

[0013] Optionally, defining the loss function includes combining the Huber loss and the physical constraint loss to obtain the total loss function, and the formula is as follows: ; Among them, λ is the weight coefficient, L total is the total loss function; L Huber is the Huber loss, L mass is the physical constraint loss, and the weight coefficient gradually increases in the order of the model underfitting period, the model preliminary convergence period, and the model overfitting risk period; The formula for the physical constraint loss is: ; Among them, M pred is the total mass calculated according to the total number of predicted particles, and M measured is the total mass obtained by mass spectrometry testing.

[0014] Optionally, the joint optimization of the shape factor and density includes obtaining a physically reasonable shape factor and material density step by step under the physical gradient constraint volume residual backpropagation of the shape factor and the iterative probability matrix of the material density EM algorithm.

[0015] The beneficial effects of the high-throughput microplastic mass spectrometry detection device and method for multi-dimensional feature synchronous detection provided by this application are as follows: 1. The four-dimensional information of the microplastic particle size, quantity, type, and mass in the sample is synchronously obtained through the fast image acquisition and analysis module and the mass spectrometry detection module, which has the advantages of simple and fast operation. Based on the characteristics of the local high temperature and fast pyrolysis rate of the electromagnetic thermal cracking device, the fast image acquisition and analysis module is installed directly above the electromagnetic thermal cracking module and cooperates with the shielding gas introduced from the top, solving the problems of slow heating rate, intolerance to high temperature of the microscope lens, and contamination of the window by pyrolysis products during the operation of the conventional resistance thermal cracking method, and can meet the requirements of high-throughput testing of large sample sizes; 2. During the formation of microplastics, there are significant differences in the particle size distributions of different polymers, which are mainly determined by the chemical structure, physical properties, environmental degradation mechanism of the material, and the morphology of the initial product. Through this device and method, the four-dimensional information of the microplastic particle size, quantity, type, and mass can be synchronously obtained, solving the problem of poor comparability of experimental results after multi-channel data acquisition, and improving the consistency and accuracy of microplastic analysis and detection; 3. It can be applied to the synchronous acquisition of multi-dimensional features of large sample sizes of microplastics in water, air, and soil, and is also applicable to the field of environmental toxicology analysis, synchronously evaluating the importance of particle size and mass in toxicology. Due to comprehensive information acquisition, this application also provides a consistent solution for the establishment of microplastic-related reference materials and standard methods. Description of the Drawings

[0016] In order to more clearly illustrate the technical solutions in the embodiments of this application, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art.

[0017] Figure 1 is a schematic structural diagram of the high-throughput microplastic mass spectrometry detection device for multi-dimensional feature synchronous testing provided by the embodiment of this application.

[0018] Figure 2 is a diagram of the microscope and mass spectrometry data alignment strategy provided by the embodiment of this application; Figure 3 is a flowchart of microplastic shape density calibration provided by the embodiment of this application. Figure 4 This is the machine learning training flow chart for analyzing the mass - quantity relationship of microplastics provided by the embodiments of the present application.

[0019] Explanation of reference numerals: 1 - Fast image acquisition and analysis module; 2 - Electromagnetic thermal cracking quartz chamber; 3 - Carrier gas inlet; 4 - Quartz stage; 5 - Tungsten sheet; 6 - Sample; 7 - Mass spectrometry detection module; 8 - Electromagnetic induction coil; 9 - Fixing part; 10 - Stage support; 11 - Bottom plate of thermal cracking quartz chamber; 12 - Visual window; 13 - Visual window shielding gas purge port; 14 - Split air extraction pump. Detailed implementation manners

[0020] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present application more clear and understandable, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0021] Embodiment 1 The object of the present invention is to provide a device capable of synchronously and high - throughput testing the above - mentioned four - dimensional characteristics of microplastics for the technical problem that it is difficult to synchronously and quickly capture multi - dimensional characteristics such as the quantity, type, particle size, and mass of microplastics in the prior art.

[0022] As Figure 1 shown, the high - throughput mass spectrometry detection device for synchronous testing of multi - dimensional characteristics of microplastics mainly consists of a fast image acquisition and analysis module 1, a gas path transmission module, an electromagnetic thermal cracking module, a filtration and purification module, a mass spectrometry detection module 7 and a data analysis module. The gas path transmission module, the electromagnetic thermal cracking module, the filtration and purification module and the mass spectrometry detection module 7 are connected in sequence through quartz pipelines, and each module is hermetically connected through high - temperature - resistant quartz pipelines to form a closed - loop analysis system. The fast image acquisition and analysis module 1 is arranged directly above the electromagnetic thermal cracking module and separated by the visual window 12. The visual window shielding gas purge port 13 is arranged below the visual window 12, and the electromagnetic thermal cracking module is connected with a split air extraction pump 14.

[0023] In this embodiment, the fast image acquisition and analysis module 1 includes a high - resolution microscope, a fast CCD camera (frame rate 600fps) and an image processing device. The image processing device, based on the image processing program Open CV installed in it, identifies the proportion of the number of surface microplastics and the particle size distribution in the acquired images.

[0024] To ensure the high - vacuum condition of the mass spectrometry, the gas path transmission module uses high - purity argon as the carrier gas source, controls the gas flow at 0.15 L / min through a control valve, needs to introduce the carrier gas for 1 min before the test starts, and simultaneously turns on the split air extraction pump 14 to remove the impurity gas in the system and maintain the vacuum state.

[0025] The electromagnetic thermal cracking module is composed of an electromagnetic thermal cracking quartz cavity 2 with an electromagnetic induction coil 8 tightly wound around its outer wall. The electromagnetic thermal cracking quartz cavity 2 is fixed on the thermal cracking quartz cavity bottom plate 11 through a fixing member 9. The two ends of the electromagnetic induction coil 8 are connected to a programmed temperature controller. A quartz sample stage 4 and a sample stage support 10 are arranged inside the electromagnetic thermal cracking module. During the heating process, the sample 6 is placed on a tungsten sheet 5, and the tungsten sheet 5 and the sample 6 are jointly placed on the quartz sample stage 4. The sample stage support 10 has an automatic lifting function, and adjusts the distance between the thermal cracking quartz cavity 2 and the microscope lens as needed to form an image within the focal length. The sample stage support 10 is installed above the thermal cracking quartz cavity bottom plate 11.

[0026] The temperature control implements dynamic PID regulation, achieving a response speed of 0.1 s within the range of ±5 °C of the target temperature. The temperature is measured by a thermocouple temperature sensor, and the probe of this sensor is in close contact with the tungsten sheet 5 below the tungsten sheet 5.

[0027] The filtration and purification module includes quartz fiber and metal filters. The target gas pyrolyzed is purified through the quartz fiber and metal filters, and after being separated by a shunt air extraction pump 14, it is transported to the mass spectrometry detection module 7 through a capillary for rapid analysis.

[0028] The mass spectrometry detection module 7 is a bench-top quadrupole mass spectrometer in this application, equipped with the ability to detect volatile gases. As a feasible implementation method, a portable quadrupole mass spectrometer can be used for on-site detection.

[0029] The working process of the system is as follows: S1: Place the microplastic sample 6 in the electromagnetic thermal cracking module, and collect the image of the sample 6 through the rapid image acquisition and analysis module 1; Before the sample 6 to be measured is placed into the electromagnetic thermal cracking module, it is first placed on the tungsten sheet 5, and the particle size distribution parameters are quickly captured by the rapid image acquisition and analysis module 1 installed within the appropriate focal length directly above. S2: Provide a vacuum environment for the electromagnetic thermal cracking module through the shunt air extraction pump 14, and then pyrolyze the sample 6 through the electromagnetic thermal cracking module. The gas path transmission module provides carrier gas to the electromagnetic thermal cracking module. The carrier gas carries the microplastic pyrolysis gas products, which are purified through the filtration and purification module and then transported into the mass spectrometry detection module 7 for analysis. Shielding gas is purged into the visible window through the window shielding gas purge port 13. The high-purity argon gas source is precisely controlled by a flow regulating device and is sent to the thermal cracking quartz cavity 2 at the carrier gas inlet 3. The electromagnetic induction coil 8 operates to pyrolyze the microplastic sample 6 on the tungsten sheet 5 into small molecule products. The pyrolysis products are purified by passing through the metal filter under the drive of the carrier gas and finally enter the mass spectrometry detection module 7 for real-time analysis, thus completing the detection of the quality and type of microplastics. S3: The fast image acquisition and analysis module 1 provides the surface microplastic particle size distribution information for the data analysis module, and the mass spectrometry detection module 7 provides the microplastic mass and type information for the data analysis module. The data analysis module physically and reasonably integrates the microplastic mass, type information obtained by mass spectrometry testing with the particle size distribution and quantity information obtained by fast image analysis through a multi-parameter machine learning model to achieve synchronous acquisition of multi-dimensional parameters of the whole machine, and analyzes through the microplastic multi-parameter machine learning model.

[0030] Example 2 A high-throughput microplastic mass spectrometry detection method for synchronous detection of multi-dimensional features. The key core in the process of obtaining four-dimensional information of microplastic quantity, particle size, type, and mass concentration is to realize the conversion between the microplastic particle size distribution (implied quantity information) obtained by image recognition and the microplastic mass information obtained by mass spectrometry testing. The implementation steps are as follows: The construction of the microplastic multi-parameter machine learning model includes: S4-1: Obtain actual test data, optimize the data through the microplastic shape density calibration program to generate reliable output data. The reliable output data forms a true value simulation training data set for machine learning through feature vector splicing, and the true value simulation training data set is divided into a training set and a test set; Data preparation and feature extraction: In the current data acquisition, there is a lack of a real data set of "particle size distribution - total mass - total number of particles - microplastic types". Therefore, it is necessary to generate machine learning training data through a physical model. Here, this application adopts a strategy of combining simulation data with experimental data, that is, using known particle size distribution parameters for modeling, and then generating simulation data with the same distribution to meet the training requirements of machine learning.

[0031] Obtain actual test data: Generate training data using microplastic standards and actual samples through a high-throughput microplastic mass spectrometry detection device with synchronous detection of multi-dimensional characteristics (in this embodiment, only microsphere standards are taken as an example, and other sheet-like, fibrous, and various shaped mixtures can also prepare data through this process). The training data includes four-dimensional information of microplastic particle size, quantity, type, and mass. In this embodiment, commercially available polyethylene, polypropylene, polystyrene, etc. (adjust the types according to actual needs) microsphere standards with different particle sizes are purchased, and the particle sizes are 300μm, 500μm, 800μm, 1mm, 2mm, 3mm, 4mm, 5mm, etc. (can be adjusted according to actual needs). Orthogonally mix the above particles according to different mass ratios to prepare mixed microplastic standards, and pay attention to drawing up a mixing ratio table in advance. From this table, we can calculate the theoretical number of particles of a certain plastic according to the density, mass, and diameter of the microplastic standard, and the sum is the total theoretical number of particles. Although the above microplastic particle size is the specific particle size mark of the merchant, its actual particle size is not completely uniform and often shows a certain particle size distribution. Therefore, such natural standard particle size mean and deviation are exactly what we need in the process of obtaining actual data. Use the fast image acquisition and analysis module to test its actual particle size distribution and record the histogram data, and use the mass spectrometry detection module to test and obtain the mass M of a certain plastic k , and obtain the actual total number of particles N through the following physical formula T .

[0032] ; In the formula, is the density of K types of microplastics; h is the shape factor, which is 2 / 3 for nearly spherical here; d i is the particle size of the i-th bucket. In this embodiment, add a ±5% random perturbation to d i , and add a ±2% random error to the total mass M, and complete data augmentation based on actual data with this as the measurement noise. As a feasible implementation method, the random error of d i depends on the microscope error, and the mass M of a certain plastic k depends on the mass spectrometry instrument test error, Microscope error calculation: Independently measure the same sample 30 times, calculate the mean and standard deviation of d i in each interval, and then calculate the relative standard deviation RSD: RSD = standard deviation / mean × 100%.

[0033] Take the maximum RSD as the perturbation range. If the average RSD is 3%, ±5% can be set to cover extreme cases.

[0034] Mass spectrometry instrument test error calculation: Measure a standard sample of known mass 30 times, calculate the 95% confidence interval, and determine the relative error range.

[0035] The above-obtained standard sample test data and supplementary data thus constitute a data set for operating the following process.

[0036] Overall operation process: As Figure 2 shown, the mass and species data from the mass spectrometry analysis module and the particle size distribution histogram data from the microscopic image analysis module are integrated and aligned through the following process. The microscope captures the video stream in real time through the GigE Vision interface, encapsulates the OpenCV image processing pipeline using Docker containers, performs image segmentation and generates the particle size distribution histogram, serializes the processing results into JSON format, and pushes them to the message bus through Kafka; the mass spectrometer outputs the raw data through the serial port, which is processed in real time by the parsing service written in Go language, extracts the mass spectrometry peak information based on the Sphinx protocol, calculates the total mass of the polymer, and pushes the data to the message bus through gRPC. At the data alignment layer, a two-stream connection (Window Join) between the microscope and the mass spectrometer is implemented based on Apache Flink, with the sample ID as the primary key, and the microscope and mass spectrometry data are aligned within a 3-minute time window.

[0037] The above-obtained standard sample data set forms a preliminary prediction model for machine learning training offline after simple physical judgment constraints. In order to comprehensively consider the quantitative loss caused by the actual shape and density differences of microplastics, an optimization system based on traditional physical constraints is designed during the actual sample test process. The optimized data continues to update the preliminary machine learning prediction model to achieve the training of the model's generalization ability and make it applicable to most application scenarios.

[0038] As Figure 3 and Figure 4 shown, the optimization system based on traditional physical constraints calculates the image volume (based on the particle size distribution histogram, empirical shape factor, and density distribution) and the mass spectrometry volume (based on the mass and density of a certain type of microplastic) respectively based on the microscope and mass spectrometry data, and compares the volume differences between the two under the traditional physical model constraints shown below.

[0039] Volume calculation V based on mass spectrometry data M : ; Volume calculation V based on image data E : ; In the formula, V M is the volume of K types of microplastics calculated by mass spectrometry test; VE is the volume of the expected K types of microplastics in image recognition; M k is the mass of the K types of microplastics obtained by mass spectrometry; ρ k is the density of the K types of microplastics, which is a fixed value; n i is the proportion of the number of particles in each particle size range; d i is the diameter of the particle size range; h is the shape factor; P ki is the proportion of the K types of plastics in the i-th particle size range; 1 / ρ k is the volume conversion factor, which converts the proportion of the number into the proportion of the volume. Here, the key initial assumption is that the single-particle volumes of all particles in the i-th particle size segment are the same, but the number of particles of different microplastics is different.

[0040] In this embodiment, if ΔV > 5%, the next-level joint optimization and calibration process of the shape factor h and the material density ρ based on the traditional physical model is triggered (this is the traditional optimization process); under the physical gradient constraint of the shape factor (backpropagation of the volume residual ∂ΔV / ∂h) and the EM algorithm probability matrix iteration of the material density, the physically reasonable shape factor and material density are obtained step by step. The range of the shape factor is limited to [0.15, 3.0]. After the density calibration, the convergence determination is performed. If Max diff < 0.1%, the calibrated probability matrix is output for physical constraint verification: V calib = ΣΣ((1 / ρ k ) × n i × P ki × v i ); In the formula, V Calib is the calibrated volume, v i is the particle volume in the i-th particle size range; If |V calib - V M | < 1%, the final credible output is output.

[0041] The final credible output includes: the particle size range distribution (20 intervals and statistical verification data (μ, σ)), the morphological characteristic parameters (h), the material density distribution (ρ) of the microplastic particle characteristic parameters, the mass parameter (M), and the total number of particles (N) obtained by the final operation. The above credible output data are spliced by feature vectors to form a true value simulation training data set for machine learning, and its optimization process is as Figure 3 shown.

[0042] S4-2: Establish a neural network model including an input layer, a hidden layer, physical constraint fusion, and an output layer; This application uses a deep neural network (DNN) to capture the non-linear relationships among the total mass, particle size distribution parameters, and particle number characteristics, and to make effective predictions. The neural network model design structure extracts the high-dimensional representation of the input features by gradually increasing the dimension of the hidden layer, and finally predicts the target variable (i.e., the total number of particles N) through the output layer. The implementation steps are as follows: Input layer: That is, the input feature dimension. The particle size interval distribution (20 intervals and statistical verification data (μ, σ)), morphological feature parameters (h), microplastic particle characteristic parameters of the material density distribution (ρ), and mass parameter (M), a total of 25 features, are used as the input of the neural network; Hidden layer: As the core part of the feature encoder, it is used to extract the high-dimensional representation of the input feature dimension. Here, the 25 dimensions of the input layer are gradually transformed through three fully connected layers (25→128→256→128). After each layer, a batch normalization layer is connected to stabilize the training process, then a ReLU activation function is connected to introduce non-linearity, and finally a Dropout layer is connected to randomly discard some neurons to prevent overfitting.

[0043] Physical constraint fusion: The purpose of physical constraint fusion is to combine the total mass measured by mass spectrometry (one of the input features) with the hidden layer features of the model to ensure that the prediction results of the model (total number of particles and microplastic types) are consistent with physical laws. Reduce the degrees of freedom of the model, avoid prediction results that do not conform to physical laws, ensure that the predicted total number of particles should match the total mass measured by mass spectrometry, and at the same time, as a regularization method, reduce overfitting and improve the performance of the model on unknown data. This physical constraint fusion process mainly maps the total mass obtained from mass spectrometry tests to the same dimension as the hidden layer features through 1 fully connected layer (here 1→128); then the mapped total mass feature and the hidden layer feature are multiplied element by element to achieve physical constraint fusion; the fused feature is passed to the output layer to predict the total number of particles.

[0044] Output layer: After passing through the fully connected layer, the input dimension 128 is changed to the output dimension of 1 (128→1), and the high-dimensional features after the physical constraint fusion in the previous step are mapped to the final target variable, the total number of particles.

[0045] S4-3: The neural network model is trained through the training set by initializing the model, instantiating the model, defining the loss function and the training loop and optimization process; Initialize the model: According to the designed neural network architecture, use the deep learning framework to define the neural network architecture, and perform parameter initialization through the default weight initialization method to ensure the stability of the model in the initial stage of training.

[0046] Model instantiation: Call the defined neural network architecture to generate a specific model object, load it onto the specified computing device, and perform subsequent training and inference on this device.

[0047] Define the loss function: Combine the Huber loss and the physical constraint loss to ensure that the prediction results are both accurate and compliant with physical laws. The formula is as follows: Huber loss: Used to measure the error between the total number of particles predicted by the model and the true value. The Huber loss behaves as the mean squared error (MSE) when the error is small and as the mean absolute error (MAE) when the error is large. The formula is:

[0048] where y is the true total number of particles, is the total number of particles predicted by the model, and δ is the error threshold (usually taken as 1).

[0049] Physical constraint loss: Used to ensure that the total number of particles predicted by the model is consistent with the total mass measured by mass spectrometry. The formula is: ; where M pred is the total mass calculated based on the predicted total number of particles, and M measured is the total mass obtained from mass spectrometry testing.

[0050] Total loss function: Combine the Huber loss and the physical constraint loss to obtain the total loss function. The formula is as follows: ; where λ is the weight coefficient used to balance the influence of the two parts of the loss. L total is the total loss function; L Huber is the Huber loss, and L mass is the physical constraint loss. The weight coefficient gradually increases in the order of the model underfitting period, the model preliminary convergence period, and the model overfitting risk period. Use dynamic adjustment, start training with a low weight, and gradually increase the importance of physical constraints to ensure that accuracy is the main focus in the early stage and consistency is enhanced in the later stage.

[0051] Training Loop and Optimization: The training data is divided into small batches and loaded onto a computing device (such as a GPU), and then multiple rounds of iterative training are carried out. Each iteration includes: forward propagation (obtaining predicted values from input features), calculating the loss (calculating the total error by combining the Huber loss and the physical constraint loss), backpropagation (calculating gradients and updating model parameters, continuously adjusting the weights and biases of the network to minimize the loss function), and recording the loss (saving the loss values of each batch for analysis and evaluation). After each round of training, the parameters of the current model (such as weights and biases) are saved for subsequent verification or deployment. This process gradually improves the prediction accuracy by repeatedly optimizing the model parameters. In model training, the AdamW optimizer is used to update the parameters to efficiently adjust the parameter update pace and prevent overfitting.

[0052] S4-4: Validate the neural network model through the test set and evaluate the model performance.

[0053] Select samples from the test set as the input for model validation. After inputting them into the trained model, forward propagation is performed to obtain the predicted total number of particles. According to the error range recorded during the training process, calculate the confidence interval of the predicted value and conventional regression metrics such as R 2 (coefficient of determination), MAE (mean absolute error), MSE (mean squared error), etc. Finally, compare the predicted values with the true values to evaluate the performance of the model.

[0054] Overall Workflow As Figure 4 shown, the data obtained from the microscope and mass spectrometry are preliminarily processed, and then preliminary volume constraints are imposed. Particles that meet the constraint conditions enter the machine learning prediction channel to obtain the output value of the final total number of particles N; data that do not meet the constraint conditions enter the traditional physical constraint optimization channel. After joint optimization of the shape factor and density, a reliable output of the multi-dimensional characteristics of microplastics is formed. This output is spliced into a feature vector and continues to enter the machine learning prediction channel to update the machine learning model in real time; samples that do not meet the standards in the machine learning prediction channel automatically flow back to the traditional physical constraint optimization channel to generate new training data, and periodic (weekly / monthly frequency) model training and real-time updates are carried out in combination with the traditional optimization channel. In this way, the historical training model can be called in real time for real-time updates during each detection process, enhancing the practicality of the model.

[0055] In Figure 4 the shown process, the data output by the machine learning prediction channel is verified through mass conservation verification and volume conservation verification, and finally the result of the total number of particles is output. Data that do not pass the volume conservation verification re-enter the traditional physical constraint optimization channel, and the data is optimized to generate reliable output data.

[0056] Using this method to fuse image and mass spectrometry data for microplastic multi-dimensional feature prediction, calibration of mass spectrometry data to image data is achieved at the data level, and supervision of machine learning models by traditional physical optimization models is realized at the method level. The dynamic complementarity between traditional physical optimization and machine learning prediction technologies ensures that the microplastic mass-quantity conversion process has the dual advantages of physical rationality and continuous improvement of prediction accuracy.

[0057] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A high-throughput mass spectrometry detection device for microplastics with multi-dimensional feature synchronous testing, characterized in that: include: A rapid image acquisition and analysis module, a gas transmission module, an electromagnetic thermal cracking module, a filtering and purification module, a mass spectrometry detection module and a data analysis module, wherein the gas transmission module, the electromagnetic thermal cracking module, the filtering and purification module and the mass spectrometry detection module are sequentially connected through a quartz pipeline, the rapid image acquisition and analysis module is arranged directly above the electromagnetic thermal cracking module and is separated by a visible window, a window shielding gas purge port is arranged below the visible window, and the electromagnetic thermal cracking module is connected to a diversion vacuum pump.

2. The high-throughput mass spectrometry detection device for microplastics with multi-dimensional feature synchronous testing as claimed in claim 1, characterized in that: The rapid image acquisition and analysis module includes a microscope, a camera and an image processing device; the gas transmission module includes a carrier gas source and a control valve; the electromagnetic thermal cracking module includes a thermal cracking quartz chamber, an electromagnetic induction coil, a quartz stage and a stage support; the filtration and purification module includes a quartz fiber and a metal filter; the mass spectrometry detection module is equipped with a mass spectrometry detection and analysis device with volatile gas detection capabilities.

3. A high-throughput mass spectrometry detection method for microplastics with multi-dimensional feature synchronous testing, characterized by: The high-throughput microplastic mass spectrometry detection device using the multi-dimensional feature synchronous detection as described in any one of claims 1-2 comprises the following steps: S1: Place microplastic samples in the electromagnetic pyrolysis module and collect sample images through the rapid image acquisition and analysis module; S2: Provide a vacuum environment for the electromagnetic pyrolysis module through a split flow vacuum pump, and then pyrolyze the sample through the electromagnetic pyrolysis module. The gas transmission module provides carrier gas to the electromagnetic pyrolysis module. The microplastic pyrolysis gas product is purified by the filtration and purification module and then transported to the mass spectrometry detection module for analysis. The shielding gas is input through the window shielding gas purge port to purge the visual window. S3: The rapid image acquisition and analysis module provides the data analysis module with the particle size distribution information of surface microplastics, and the mass spectrometry detection module provides the data analysis module with the overall microplastic quality and type information. The data analysis module performs analysis using a microplastic multi-parameter machine learning model.

4. The high-throughput mass spectrometry detection method for microplastics with multi-dimensional feature synchronous testing as claimed in claim 3, characterized in that: The construction of the microplastic multi-parameter machine learning model includes: S4-1: Acquire actual test data, optimize the data through a microplastic shape density calibration program to generate credible output data, concatenate the credible output data through feature vectors to form a true value simulation training data set for machine learning, and divide the true value simulation training data set into a training set and a test set; S4-2: Establish a neural network model including input layer, hidden layer, physical constraint fusion layer and output layer; S4-3: Train the neural network model through the training set, initializing the model, instantiating the model, defining the loss function, and training loop and optimization process; S4-4: Validate the neural network model through the test set and evaluate the model performance.

5. The high-throughput mass spectrometry detection method for microplastics with multi-dimensional feature synchronous testing as claimed in claim 4, characterized in that: The actual test data acquisition includes: using microplastic standards and actual samples, and using a high-throughput microplastic mass spectrometry detection device with multi-dimensional feature synchronous detection to obtain the quality, type, quantity and particle size information of the microplastics.

6. The high-throughput mass spectrometry detection method for microplastics with multi-dimensional feature synchronous testing as claimed in claim 4, characterized in that: The data optimization to generate credible output data includes respectively calculating the image volume and density distribution based on the particle size distribution histogram and the empirical shape factor and the mass spectrum volume based on the mass and density of the microplastics, performing volume constraints, and directly using the data that meets the constraints as credible output data to enter the machine learning prediction channel to obtain the final output value of the total number of particles; The data that does not meet the constraints enters the volume constraint optimization channel, and after the joint optimization of shape factor and density, reliable output data is formed.

7. The high-throughput mass spectrometry detection method for microplastics with multi-dimensional feature synchronous testing as claimed in claim 4, characterized in that: The mass spectrum volume is calculated according to the following formula: ； Where V M The volume of K types of microplastics was calculated for mass spectrometry, M k is the mass of K type microplastics obtained by mass spectrometry; ρ k is the density of K types of microplastics; The image volume is calculated according to the following formula: ； Where V E is the volume of K types of microplastics expected to be obtained in image recognition; ρ k is the density of K types of microplastics, which is a fixed value; n i is the proportion of particles in each particle size range; d i The diameter of the particle size interval; h is the shape factor; P ki is the proportion of K types of plastics in particle size range i; 1 / ρ k is the volume conversion factor, which converts the quantity percentage into the volume percentage.

8. The high-throughput mass spectrometry detection method for microplastics with multi-dimensional feature synchronous testing as claimed in claim 4, characterized in that: The physical constraint fusion includes mapping the total mass of microplastics obtained from the mass spectrometry test to the same dimension as the hidden layer features through a fully connected layer, and multiplying the mapped total mass features by the hidden layer features element by element to achieve physical constraint fusion; the fused features are passed to the output layer to predict the amount of microplastics.

9. The high-throughput mass spectrometry detection method for microplastics with multi-dimensional feature synchronous testing as claimed in claim 4, characterized in that: The definition of the loss function includes combining the Huber loss and the physical constraint loss to obtain the total loss function, which is as follows: ； Among them, λ is the weight coefficient, L total is the total loss function; L Huber is the Huber loss, L mass For physical constraint loss, the weight coefficient is gradually increased in the order of model underfitting period, model initial convergence period and model overfitting risk period; The physical constraint loss formula is: ； Among them, M pred is the total mass calculated based on the predicted total number of particles, M measured is the total mass obtained by mass spectrometry test.

10. The high-throughput mass spectrometry detection method for microplastics with multi-dimensional feature synchronous testing as claimed in claim 6, characterized in that: The joint optimization of the shape factor and density includes obtaining the shape factor and material density that conform to physical rationality step by step under the shape factor physical gradient constraint volume residual back propagation and material density EM algorithm probability matrix iteration.

Images