Electronic waste disassembly pollutant screening method based on non-targeting technology

Abstract

The invention relates to the technical field of environment monitoring and pollutant traceability, and discloses an electronic waste disassembly pollutant screening method based on a non-targeted technology, and the method comprises the steps: employing an Internet of Things sensing network module to monitor environment and working condition characteristics, calculating a real-time pollution load index, and triggering sampling when the index continuously exceeds a threshold value; the integrated on-line sampling module is used for intercepting particulate matters by utilizing an inertia separation structure through adjusting airflow, and capturing volatile organic compounds by utilizing an adsorption medium; introducing the sample into a mobile high-resolution analysis module through flash evaporation thermal desorption, and acquiring mass spectrum data by using a soft ionization source; the intelligent identification and traceability module constructs a deep neural network model based on physical constraints, processes data through non-negative projection and sparse regularization, and analyzes the contribution rate and orientation probability of a disassembly process in combination with wind field information. Accurate capture and space-time traceability of complex pollutants in the electronic waste disassembling process are achieved.

Description

Technical Field

[0001] This invention relates to the field of environmental monitoring and pollutant source tracing technology, specifically to a method for screening pollutants from the dismantling of electronic waste based on non-targeting technology. Background Technology

[0002] The electronic waste dismantling industry is a typical source of complex pollution emissions. Its operations involve multiple processes such as crushing, heat treatment, and chemical leaching, releasing mixed pollutants including heavy metals, polybrominated diphenyl ethers (PBDEs), polycyclic aromatic hydrocarbons (PAHs), and a large number of unknown pyrolysis products. These pollutants are complex in composition, highly toxic, and persistent in the environment, posing a potential threat to the surrounding ecosystem and human health. Therefore, establishing efficient methods for pollutant screening and source tracing is of great significance for environmental supervision and risk prevention in industrial parks.

[0003] Current monitoring technologies for pollutants from electronic waste dismantling primarily rely on fixed-point offline sampling and laboratory analysis. This approach typically involves long-term integrated sampling according to a pre-set schedule. However, electronic waste dismantling operations are often characterized by significant intermittent and sporadic events. Fixed sampling cycles struggle to capture instantaneous high-concentration emission events, and key pollutant characteristics are easily diluted by background air, failing to accurately reflect the pollution load during peak operation periods. Furthermore, offline analysis suffers from significant time lags, making it difficult to meet the environmental regulatory requirements for rapid response to sudden pollution incidents.

[0004] In terms of analysis, detection, and source tracing, existing technologies mostly employ targeted analysis strategies, i.e., quantitative detection of known specific pollutants. However, electronic waste generates a large number of secondary pollutants or thermal degradation products with unknown structures during thermal dismantling. Focusing only on known targets can lead to the omission of a large number of potentially toxic and hazardous substances, making it impossible to comprehensively assess environmental risks. Furthermore, traditional data analysis methods, when processing the massive amounts of data generated by high-resolution mass spectrometry, often use receptor models such as positive definite matrix factorization for source analysis. These models often lack strict physical constraints during mathematical solutions, easily resulting in negative source contribution values ​​in the calculation results, which do not conform to objective physical laws. They also struggle to handle the sparsity problem of high-dimensional data, leading to a discrepancy between the analyzed pollution source fingerprint and the actual dismantling process, making it impossible to accurately determine the specific generation stage and spatial location of pollutants. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a non-targeted technology-based method for screening pollutants from electronic waste dismantling. This method solves the problems of existing electronic waste dismantling pollutant monitoring technologies, such as the difficulty in capturing instantaneous emissions through fixed-point offline sampling, the tendency to miss unknown secondary pollutants through targeted analysis, and the lack of physical constraints in traditional source apportionment models leading to insufficient accuracy in source tracing.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solution: a method for screening pollutants from the dismantling of electronic waste based on non-targeting technology, which mainly includes four stages: environmental perception triggering, integrated sampling, analysis and detection, and intelligent source tracing.

[0007] During the environmental sensing triggering phase, environmental parameters, meteorological parameters, location information, operating conditions and material characteristics are monitored in real time through the Internet of Things sensing network module. The system has set threshold values ​​for characteristic indicators. When the monitored characteristic indicator values ​​exceed the preset threshold, a start signal is sent to the integrated online sampling module to achieve targeted capture of high pollution periods or specific operating conditions.

[0008] During the integrated sampling phase, the integrated online sampling module starts the sampling pump after receiving the start signal. The module uses an inertial separation structure to classify and retain particulate matter, causing the particulate matter to deposit on the filter membrane. At the same time, it uses the adsorption medium to capture volatile organic compounds in the airflow. After sampling, the thermal desorption program is started to heat the filter membrane and adsorption medium, causing the captured pollutants to desorb and vaporize. The sample is then introduced into the mobile high-resolution analysis module.

[0009] During the analysis and detection phase, the mobile high-resolution analysis module uses a soft ionization source to acquire mass spectrometry data of the sample in full scan mode, forming a raw data stream containing rich chemical information.

[0010] During the intelligent source tracing stage, the intelligent identification and source tracing module processes the raw data stream to generate feature vectors and compares them with the database to determine the types of pollutants. Subsequently, the module combines the determined types of pollutants, the pollution source fingerprint database, and operating condition information, and applies chemometric algorithms to calculate the contribution rate of different dismantling processes or materials, and finally generates an analysis report.

[0011] As a preferred technical solution, the IoT sensing network module is equipped with a particulate matter sensor based on the principle of light scattering, a photoionization detector, and a material fingerprint recognition unit. This module calculates the real-time pollution load index through an edge computing gateway, which serves as the basis for triggering sampling. The calculation method of the real-time pollution load index is as follows: the ratio of particulate matter mass concentration to a preset particulate matter concentration benchmark threshold, the ratio of volatile organic compound concentration to a preset volatile organic compound concentration benchmark threshold, and the material spectral feature matching degree are multiplied by the corresponding dimensionless weighting coefficients and then weighted and summed.

[0012] Furthermore, to avoid false triggering, the system has set specific triggering logic: the edge computing gateway must determine that the real-time pollution load index is greater than the trigger threshold and that the duration of this state exceeds the set minimum confirmation time before it can send a start signal.

[0013] In the specific implementation of the sampling structure, the integrated online sampling module controls the average airflow velocity at the nozzle outlet by adjusting the pumping flow rate of the downstream sampling pump. This velocity control prevents coarse particles with an aerodynamic diameter larger than the cutting diameter from entering the receiving tube, while fine particles smaller than the cutting diameter are deflected by the main airflow and guided through the built-in filter membrane. The adsorption medium is set downstream of the filter membrane, and weakly polar adsorbent and strongly polar adsorbent are sequentially filled along the airflow direction to achieve broad-spectrum capture of organic compounds of different polarities.

[0014] The thermal desorption process consists of two steps: purging and heating. First, the flow path control unit drives the valve assembly to establish a purging circuit using inert gas, removing residual moisture and oxygen from the adsorption medium. Then, the heating unit executes a flash heating strategy, controlling the heating rate to rapidly increase the temperature. During sample transfer, the inner wall of the transfer pipeline between the mobile high-resolution analysis module and the integrated online sampling module is made of inert material, and the outside is wrapped with a heat tracing cable and an insulation layer. A PID controller maintains the pipe wall temperature at a constant high temperature to prevent the condensation and residue of high-boiling-point components.

[0015] To ensure the quality of the detection data, the mobile high-resolution analysis module uses a steel wire rope vibration isolation system to isolate the built-in mass spectrometer, reducing environmental vibration interference during mobile monitoring. At the same time, the module periodically injects internal standard compounds with known precise mass numbers through an independent reference channel, and reverse-calibrates the mass axis coefficient based on the measured flight time of the internal standard compounds to ensure the accuracy of the mass numbers.

[0016] In terms of data processing algorithms, the intelligent identification and source tracing module constructs a deep neural network model based on physical constraints. This model includes an encoder and a decoder structure: the encoder compresses the normalized concentration matrix constructed based on feature vectors into the hidden layer, and the hidden layer uses the modified linear unit as the activation function; the decoder reconstructs the hidden layer features back to the original dimension, and its connection weights directly correspond to the source component spectrum. When updating the weights during backpropagation, the projection operator is used to reset the updated negative weights to zero, so as to conform to the physical law that pollutant concentration is non-negative.

[0017] The training objective of this deep neural network model is achieved by minimizing the loss function. The loss function consists of a weighted reconstruction error term and a sparse regularization term. The weighted reconstruction error term contains the squared difference between the measured concentration and the network's reconstructed output value, weighted using measurement uncertainty. The sparse regularization term contains the product of the L1 regularization coefficient and the source contribution term corresponding to the hidden layer features, inducing sparse solutions.

[0018] Before calculating the contribution rate, the system performs a data cleaning step. The intelligent identification and source tracing module uses the timestamp of the mass spectrometry data as a benchmark to upsample and reconstruct the meteorological and location data contained in the environmental parameters, and identifies and removes data from calm wind periods with wind speeds less than the set value. At the same time, it performs pre-analysis of the concentration matrix through singular value decomposition and determines the number of potential pollution sources based on the cumulative contribution rate of eigenvalues.

[0019] In addition, this method also realizes the spatial orientation analysis of pollution sources. After the intelligent identification and source tracing module analyzes the time contribution sequence of each pollution source, it calculates the conditional bivariate probability function by combining the synchronous wind field data contained in the environmental parameters, and maps the contribution in the time dimension to the orientation in the spatial dimension. In order to improve the statistical significance, when the total number of samples in a certain wind direction sector is less than the minimum statistical threshold, the probability value in that wind direction sector is forced to be zero. Finally, the probability radar chart that intuitively displays the orientation of each pollution source is output.

[0020] This invention provides a method for screening pollutants from the dismantling of electronic waste based on non-targeting technology. It has the following beneficial effects:

[0021] 1. This invention uses an IoT sensing network module to monitor multi-dimensional environmental parameters and operating conditions in real time. It uses an edge computing gateway to calculate a real-time pollution load index that integrates particulate matter, volatile organic compounds, and material characteristics, and triggers sampling when the index continuously exceeds a threshold. This triggering mechanism is designed to address the instantaneous and intermittent nature of pollutant emissions during the dismantling of electronic waste. It achieves accurate capture of high-pollution periods or specific operating conditions, avoiding the problem that traditional fixed-frequency sampling methods may miss key emission events or collect a large amount of invalid background data, thus significantly improving the targeting and efficiency of screening.

[0022] 2. In terms of hardware, this invention adopts a structure combining inertial separation and staged adsorption, coupled with an inert high-temperature heated transmission pipeline and a flash thermal desorption procedure, to achieve simultaneous acquisition and non-destructive injection of particulate matter and volatile organic compounds. This effectively prevents condensation residue or adsorption loss of high-boiling-point organic compounds during transmission. At the same time, combined with a steel wire rope vibration isolation system designed for mobile monitoring and real-time internal standard calibration technology, it overcomes vibration interference and instrument drift during vehicle travel, ensuring that the mass spectrometry data acquired by the mobile high-resolution analysis module under complex working conditions has high fidelity and mass axis accuracy.

[0023] 3. This invention constructs a deep neural network model based on physical constraints. By introducing a nonnegative projection operator and a sparse regularization term in backpropagation, the model weights are forced to conform to the physical laws of nonnegative pollutant concentration and sparse sources, solving the problem that the analytical results of traditional black-box models lack practical physical meaning. Furthermore, by combining synchronous wind field data to calculate a conditional bivariate probability function, the time contribution sequence of each dismantling process or material is mapped to spatial orientation probability, thereby simultaneously locking the chemical fingerprint type of the pollution source and its specific spatial location, providing accurate traceability basis for refined environmental supervision of electronic waste dismantling parks. Attached Figure Description

[0024] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a schematic diagram of the system architecture of the present invention; Figure 3 This is a time-series comparison diagram of pollution source analysis according to the present invention; Figure 4 This is a comparison and verification diagram of the convergence of the algorithm of the present invention.

[0025] Among them, 100 is an integrated online sampling module; 200 is an IoT sensing network module; 300 is a mobile high-resolution analysis module; and 400 is an intelligent identification and traceability module. Detailed Implementation

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

[0027] Please see the appendix Figure 1 This invention provides a method for screening pollutants from electronic waste dismantling based on non-targeting technology, comprising the following steps: S10, the Internet of Things sensing network module 200 continuously monitors the environmental parameters and material characteristics of the dismantling workshop. When the monitored characteristic index value exceeds the preset threshold, it sends a start signal to the integrated online sampling module 100. S20, the integrated online sampling module 100 receives the start signal and turns on the sampling pump, introduces the ambient airflow into the sampling probe, and uses the internal inertial separation structure to classify and intercept particulate matter in the airflow according to aerodynamic diameter, and uses the adsorption medium to capture volatile organic compounds in the airflow. S30. After sampling, the integrated online sampling module 100 starts the thermal desorption program to program the temperature of the filter membrane and adsorption medium enriched with pollutants, so that the adsorbed organic pollutants are desorbed and vaporized. The vaporized sample is then introduced into the mobile high-resolution analysis module 300 using the carrier gas. S40, the mobile high-resolution analysis module 300 uses a soft ionization source to convert gaseous samples into ions, acquires mass spectrometry data in full scan mode, records the precise mass-to-charge ratio, isotope distribution and ion intensity information of ions in the sample, and forms a raw data stream. S50, the intelligent identification and traceability module 400 receives the raw data stream, performs baseline correction and peak extraction processing to generate a feature vector table, compares the feature vector table with the built-in electronic waste pollutant database, calculates the identification score based on the mass deviation and isotope matching degree, and determines the type of pollutant. S60, the intelligent identification and traceability module 400 retrieves the pollution source fingerprint database, combines it with the identified list of pollutant types and the operating condition information provided by the Internet of Things sensing network module 200, and uses chemometrics algorithms to calculate the contribution rate of different dismantling processes or materials to generate an analysis report.

[0028] Please see the appendix Figure 2 This invention provides a rapid non-targeted screening system for collecting and analyzing gaseous and particulate pollutants generated during the treatment of electronic waste. The system mainly includes: an integrated online sampling module 100, an Internet of Things sensing network module 200, a mobile high-resolution analysis module 300, and an intelligent identification and traceability module 400.

[0029] The integrated online sampling module 100 is installed in the work area of ​​the electronic waste dismantling workshop. It is equipped with a gas sampling channel and a particulate matter classification and capture device to simultaneously collect gaseous pollutants and particulate pollutants of different sizes from the ambient air. The module is connected to the mobile high-resolution analysis module 300 through pipelines.

[0030] The Internet of Things (IoT) sensing network module 200 is distributed at key workstations on the dismantling production line. It includes a variety of online sensors to monitor workshop environmental parameters and material characteristics. It also establishes communication connections with the integrated online sampling module 100 and the intelligent identification and traceability module 400 to send trigger control signals and auxiliary environmental data.

[0031] The mobile high-resolution analysis module 300 receives samples from the integrated online sampling module 100 through a physical interface. It integrates an ion source and a high-resolution mass spectrometer to perform ionization and full-spectrum scanning analysis on the samples and generate high-resolution mass spectrometry data.

[0032] The intelligent identification and source tracing module 400 is connected to the mobile high-resolution analysis module 300 to receive mass spectrometry data and perform feature extraction, database comparison and source apportionment calculation, and finally output pollutant screening results and source tracing report.

[0033] The electronic waste dismantling pollutant screening method of this invention is implemented based on the above-mentioned system, and its overall workflow is as follows:

[0034] The Internet of Things (IoT) sensing network module 200 continuously monitors the micro-environment of the dismantling workshop. When the monitored characteristic index value exceeds the preset threshold, the IoT sensing network module 200 sends a start signal to the integrated online sampling module 100.

[0035] After receiving the start signal, the integrated online sampling module 100 turns on the sampling pump, and the ambient airflow enters the sampling probe. Through the internal inertial separation structure, the particulate matter in the airflow is classified and intercepted according to the aerodynamic diameter, while the volatile organic compounds in the airflow are captured by the adsorption medium.

[0036] After sampling, the integrated online sampling module 100 starts the thermal desorption program. The heating device programmatically heats the filter membrane and adsorption medium enriched with pollutants, causing the adsorbed organic pollutants to desorb and vaporize. The carrier gas then introduces the vaporized sample into the mobile high-resolution analysis module 300.

[0037] The mobile high-resolution analysis module 300 uses a soft ionization source to convert gaseous samples into ions and acquires mass spectrometry data in full scan mode. This process records the precise mass-to-charge ratio, isotope distribution, and ion intensity information of all ions in the sample, forming a raw data stream containing three-dimensional information of time, mass, and intensity.

[0038] The intelligent identification and traceability module 400 receives the raw data stream, first performs baseline correction and peak extraction processing to generate a feature vector table. Then, the module compares the feature vector with the built-in electronic waste pollutant database, calculates the identification score based on the mass deviation and isotope matching degree, and determines the type of pollutant.

[0039] Finally, the intelligent identification and traceability module 400 retrieves the pollution source fingerprint database, combines it with the currently identified pollutant list and the operating condition information provided by the IoT sensing network module 200, and uses chemometric algorithms to calculate the contribution rate of different dismantling processes or materials, generating an analysis report that includes a pollutant list and source composition.

[0040] S10. Implementation of IoT Sensing and Intelligent Trigger Control:

[0041] In this embodiment, the Internet of Things sensing network module 200 performs continuous monitoring of environmental parameters and event-driven determination through sensor groups deployed at key nodes of the electronic waste dismantling line.

[0042] To accurately capture sudden and unsteady-state pollution releases during dismantling, the system constructs a multi-dimensional sensing and monitoring network at the physical level. For high-dust-generating areas such as the feed inlet of the mechanical crusher and the sorting vibrating screen, particulate matter sensors based on the principle of light scattering are configured; for volatile areas such as high-temperature baking and chemical stripping, photoionization detectors (PID) are configured; and non-contact near-infrared spectroscopy (NIR) sensors or laser-induced fluorescence (LIF) sensors are introduced as material fingerprinting units. These sensors are aggregated to the edge computing gateway via an industrial fieldbus. Given the order-of-magnitude differences in the response time constants of different types of sensors, the edge computing gateway employs zero-order hold or linear interpolation algorithms to map multi-source heterogeneous data onto a unified time axis, forming an aligned time series vector to ensure the spatiotemporal consistency of subsequent fusion computing.

[0043] After acquiring aligned environmental data, this embodiment abandons the traditional logic of triggering a response based on a single indicator exceeding its limit. Instead, it employs a multi-parameter weighted fusion algorithm to calculate the real-time pollution load, thus avoiding misjudgments caused by single sensor drift or local interference. The edge computing gateway calculates the real-time pollution load index based on a pre-set pollution load model. The physical meaning of this model lies in quantifying the overall deviation of the current workshop environment from the safety benchmark. The calculation expression is as follows: ; In the formula, The current concentration of particulate matter as measured by the light scattering sensor; The preset particulate matter concentration benchmark threshold is determined based on the static background value of the workshop or relevant occupational exposure limits, and must meet the following requirements. To avoid errors in mathematical calculations; This represents the concentration of volatile organic compounds measured by the photoionization detector at the current moment. The preset threshold for volatile organic compound concentration must also meet the following requirements. ; The material spectral feature matching degree is obtained by calculating the cosine similarity or Euclidean distance between the real-time acquired spectral vector and the pre-stored standard spectral vector of high-risk materials (such as brominated flame retardant plastics). It is used to characterize the potential chemical hazards of the currently dismantled material and the value is normalized to the [0,1] interval. These are dimensionless weighting coefficients for particulate matter, volatile organic compounds, and material characteristics, respectively, and satisfy the following conditions: The specific allocation of weights is determined based on the characteristics of the process section. For example, in the physical crushing section, the weight of particulate matter is... It was set as a dominant factor greater than 0.5, while in the baking section it was set as... Leading role.

[0044] Based on the calculated real-time pollution load index The system executes threshold discrimination logic with anti-jitter functionality. The edge computing gateway not only compares... With trigger threshold The size relationship is also determined by a time window mechanism, which requires... State duration If the minimum confirmation time set by the system is exceeded (e.g., 5 seconds), the spike noise caused by electromagnetic interference or instantaneous airflow disturbance will be filtered out. Once the judgment condition is met, the gateway will immediately generate a control command containing timestamp and trigger source type metadata, which will drive the integrated online sampling module 100 to start, realizing the transformation from passive monitoring to active capture.

[0045] S20, Simultaneous graded sampling of airborne dust from the same source:

[0046] In this embodiment, the integrated online sampling module 100 achieves simultaneous collection of gaseous pollutants and particulate pollutants in ambient air through a specific fluid dynamics flow path design, and completes real-time classification of particulate matter based on the aerodynamic equivalent diameter.

[0047] To eliminate source resolution errors caused by temporal and spatial inconsistencies in traditional split-type sampling, this embodiment constructs a series-connected integrated sampling flow path. Ambient air enters the omnidirectional sampling inlet under the stable negative pressure provided by the sampling pump. This inlet is equipped with a heating sleeve to prevent water vapor condensation in high-humidity environments. Subsequently, the airflow enters a virtual impactor assembly designed based on the principle of inertial separation. During the fluid separation process, the system regulates the trajectory of particulate matter based on the dimensionless parameter of the Stokes number; those skilled in the art know that the particle cutting diameter (d) is... 50 The aerodynamic diameter of the particles (i.e., the particle diameter when the separation efficiency reaches 50%) has a clear physical functional relationship with the airflow velocity, nozzle characteristic diameter, and particle density. In this embodiment, the system controls the average airflow velocity at the nozzle outlet by precisely adjusting the pumping flow rate of the downstream sampling pump, so that coarse particles with greater inertia (such as particles with a diameter greater than 2.5 micrometers) cannot follow the sharply deflected streamline and rush into the receiving tube; while fine particles with less inertia enter the main sampling channel with the deflection of the main airflow.

[0048] After the aforementioned inertial classification, fine particulate matter deflected by the main airflow is guided through the built-in quartz fiber or Teflon filter membrane to achieve physical interception and enrichment of the fine particulate matter. The clean airflow, stripped of particulate matter, continues downstream and enters the gaseous pollutant adsorption unit arranged in series. As a preferred embodiment, this adsorption unit employs a multi-bed adsorption tube structure, sequentially filled with a weakly polar adsorbent (such as Tenax TA) and a strongly polar adsorbent (such as Carbopack) along the airflow direction to achieve broad-spectrum capture of volatile organic compounds (VOCs) and semi-volatile organic compounds (SVOCs) with different boiling points and polarities. This layout, placing the adsorption tubes after particulate matter filtration, effectively prevents pore blockage and desorption interference caused by particulate matter deposition on the adsorbent surface.

[0049] S30. Implementation of online thermal desorption and sample import: In this embodiment, after completing the sampling task of the predetermined cycle, the integrated online sampling module 100 immediately switches to the sample pretreatment and import state. The system adopts in-situ instantaneous thermal desorption technology in conjunction with a full-process heat tracing transmission mechanism to convert the semi-volatile organic compounds and volatile organic compounds captured in the filter membrane and adsorption tube into gas phase molecular beams.

[0050] To ensure that the analytical results accurately reflect the environmental components, the flow path control unit drives the valve assembly to establish a purge circuit using inert gas with a purity greater than 99.999% to remove residual moisture and oxygen from the adsorption medium until the humidity at the flow path outlet is lower than the preset dew point value. Subsequently, the system enters the programmed temperature rise thermal desorption stage. The heating unit executes a flash heating strategy, and based on the principle of chemical reaction kinetics, the temperature rises rapidly by controlling the heating rate, thereby instantly increasing the desorption rate constant. This process forces the originally strongly adsorbed high-boiling-point pollutants (such as polybrominated diphenyl ethers) to overcome the adsorption energy barrier and be released in a concentrated manner in a very short time, thereby obtaining chromatographic peaks with narrow pulse widths in subsequent analyses and avoiding peak tailing.

[0051] The extracted gaseous pollutant molecules are then carried by the carrier gas into the transmission pipeline connecting the sampling module and the analysis module. Targeting easily condensable substances commonly found in electronic waste dismantling environments, the inner wall of the transmission pipeline is made of inert material and is wrapped with a heat tracing cable and insulation layer. The system maintains the pipe wall temperature at a constant high temperature (e.g., 280°C to 300°C) through a PID controller. This transmission design with no cold spots throughout the entire process ensures the transmission efficiency of the sample from the sampling end to the ion source inlet, realizing the reproduction of laboratory-level analytical accuracy in mobile monitoring scenarios.

[0052] S40, Implementation of mobile high-resolution full-spectrum analysis: In this embodiment, the mobile high-resolution analysis module 300 serves as the core detection unit of the system. It is responsible for receiving the gaseous sample stream from the sampling module and using the coupling of soft ionization technology and high-resolution time-of-flight mass spectrometry technology to convert the complex environmental mixture into a high-dimensional digital signal containing precise mass information.

[0053] Considering the vibration interference at the site, this embodiment implements an engineered vibration isolation modification for the precision mass spectrometer. A steel wire rope vibration isolation system with nonlinear stiffness damping characteristics is used to form a low-pass filter effect, effectively cutting off high-frequency environmental mechanical waves and ensuring that the geometrical alignment of the micron-level optical lens group inside the time-of-flight mass analyzer does not shift. After the sample is introduced into the ion source region, the system preferably adopts a soft ionization mode such as atmospheric pressure chemical ionization or low-temperature plasma to generate quasi-molecular ions, avoiding excessive fragmentation of the target object caused by the high-energy electron beam.

[0054] The generated ion beam enters the fieldless drift tube under the action of an accelerating electric field. In this physical process, the flight time of the ions is proportional to the square root of the ion mass. Since changes in ambient temperature and humidity during mobile monitoring may cause slight changes in flight distance or voltage drift, this embodiment introduces a dual-channel spray-locking mass technology for real-time parameter correction. The system periodically injects an internal standard compound with a known precise mass number through an independent reference channel, and calibrates the mass axis coefficient in reverse according to the measured flight time of the internal standard. After processing by a microchannel plate detector and a time-to-digital converter, the system finally outputs a high-resolution three-dimensional data matrix containing retention time, precise mass-to-charge ratio (m / z), and ion intensity.

[0055] S60. Implementation of pollution source apportionment and tracing model: In step S60, the system performs multi-dimensional spatiotemporal fusion of the time series of characteristic pollutant concentrations identified in the previous steps with the synchronously collected meteorological and geographical location data. This module constructs a deep neural network model based on physical constraints to adaptively calculate the contribution ratio of different pollution sources to the current environment.

[0056] For time alignment and quality control of multi-source heterogeneous data, the system uses high-frequency timestamps from mass spectrometry data as a reference and employs cubic spline interpolation to upsample and reconstruct low-frequency meteorological and location data. During this stage, the system performs data cleaning based on atmospheric diffusion theory: automatically identifying and removing wind speed data. Data on calm periods with wind speeds <0.5 m / s are used to avoid calculation divergence caused by the lack of directional significance of wind direction.

[0057] A physical constraint model based on chemical mass balance is constructed. This embodiment follows the basic assumptions of the environmental receptor model, namely that the pollutant concentration observed at the receptor point is a linear superposition of the contributions from each emission source. Before inputting the data into the neural network, the system first performs a pre-analysis of the concentration matrix through singular value decomposition, and determines the number of potential pollution sources based on the cumulative contribution rate of eigenvalues ​​(e.g., 95%). .

[0058] The construction and analysis of nonnegativity-constrained autoencoders contribute to the accurate solution of the source contribution matrix. Source component spectral matrix The system constructs a multi-layer neural network solver; this network includes an encoder and a decoder: the input layer receives the normalized concentration matrix; the encoder compresses the high-dimensional data to a dimension of 1. The hidden layer uses the modified linear unit as its activation function. Taking advantage of its zero output on the negative half-axis, the output of the hidden layer (i.e., the source contribution) is forced to be non-negative. The decoder is responsible for reconstructing the hidden layer features back to the original dimension. Its connection weights directly correspond to the source component spectrum. When updating the weights during backpropagation, the system introduces a projection operator to force all updated negative weights to zero.

[0059] The training objective of the network is to approximate the true solution by minimizing the weighted reconstruction error and the sparse regularization term. The loss function is defined as follows: ; In the formula, and These are the sample size and the number of species, respectively. This is the measured concentration; Reconstruct the network output values; To measure uncertainty, this parameter is calculated from the instrument detection limit and the relative error of the measured value, and the system sets a minimum safety threshold (e.g., 10). −9 To prevent the denominator from approaching zero; This is the L1 regularization coefficient (preferably in the range of 0.01 to 0.1). It introduces... The weighting term is introduced to eliminate the dominance of high-concentration species in the model, and the L1 regularization term is used. This is to induce the source contribution matrix. The sparsity of the receptor is consistent with the environmental physics fact that at any single moment, the contamination at a receptor point is usually contributed by only a few dominant sources.

[0060] Spatial orientation inversion based on conditional probability was performed after network training converged and the temporal contribution sequences of each pollution source were extracted. Later (here) The system combines synchronous wind field data to calculate a conditional bivariate probability function, mapping the contribution in the time dimension to the orientation in the spatial dimension. The algorithm incorporates boundary judgment logic: when the total number of samples in a certain wind direction sector is less than the minimum statistical threshold (e.g., 5), the probability value of that area is forced to be 0 to avoid small sample bias caused by data sparsity. The system finally outputs the orientation probability radar map of each pollution source, indicating the physical orientation of the suspected emission source.

[0061] Application Example – Application in a Waste Circuit Board Dismantling Workshop in South China

[0062] Application scenario description:

[0063] This embodiment selects the pyrometallurgical enrichment workshop of a large electronic waste treatment plant in South China as the application object. The workshop mainly carries out crushing, gravity sorting and low-temperature roasting pretreatment of waste circuit boards. The production environment is complex and there is a risk of compound pollution from particulate matter (PM) and volatile organic compounds (VOCs) generated by the thermal decomposition of brominated flame retardants. Moreover, the emissions have sudden and non-steady-state characteristics.

[0064] System Deployment:

[0065] Sensing layer deployment: A light scattering PM sensor is deployed at the crusher feed inlet (node ​​A) to monitor dust; a PID sensor and a miniature near-infrared spectrometer are deployed at the roasting furnace maintenance port (node ​​B) to monitor volatile organic compounds and material composition. All sensor data is encapsulated using the MQTT (Message Queuing Telemetry Transport) protocol through an industrial-grade 4G / 5G DTU module and pushed to the edge computing gateway in real time in JSON format. It also has a local caching and retransmission mechanism in the event of a network outage to ensure data integrity.

[0066] Parameter settings and formula application:

[0067] Based on the workshop background values, set the particulate matter baseline threshold. =0.5mg / m 3 .

[0068] Based on safety standards, set VOCs benchmark thresholds. =2.0ppm.

[0069] Weighting: Given the high dust generation in this section, the weighting parameter is set as follows: .

[0070] Triggering logic: The edge computing gateway calculates the real-time pollution load index according to the following formula. : ; Pollution source fingerprint database initialization: Before the system officially goes into operation, environmental background samples are collected using mobile sampling equipment under three conditions: full-load production, shutdown for maintenance, and single-process operation in the workshop. Headspace gas samples are also collected after the combustion of typical raw materials such as waste circuit boards and plastic casings. The mass-to-charge ratio distribution of characteristic ions is determined using standard laboratory analytical methods to construct a localized initial pollution source fingerprint database, which serves as the benchmark data for source apportionment calculations in step S60. Finally, the system sets trigger thresholds. =1.5, and set the minimum confirmation time. =5 seconds to prevent false alarms.

[0071] Mass spectrometer operating parameters settings: The mass scan range of the mobile TOF-MS was set to m / z 50-1200, the mass resolution was optimized to >25,000 (FWHM @m / z 200), the acquisition frequency was set to 50 spectra / s, and the electrospray ionization source voltage was set to 3.5kV to ensure effective capture of macromolecular brominated flame retardants and their degradation products.

[0072] Operation process: During a certain production shift, at 10:15 AM, due to the introduction of incompletely disassembled battery components into the feed, the crusher load suddenly increased, and the system executed the following process: T=0s: The light scattering sensor detected a PM concentration increase from 0.3 mg / m³. 3 Surge to 1.2 mg / m³ 3 .

[0073] T=2s: The PID monitor detected fluctuations in the main components of VOC concentration, and the pollution load index was calculated by substituting them into the above formula. It climbed to 1.8, exceeding the set threshold of 1.5.

[0074] From T=2s to T=7s: The system continuously monitors. Maintain a value above 1.6 to meet the 5-second anti-shake condition and confirm that there is no interference noise.

[0075] T=7s: The edge gateway issues a command, the integrated sampling module starts, and the gas path switches to sampling mode.

[0076] Subsequent processing: Sampling automatically stopped after 10 minutes. The sample was then thermally desorbed and analyzed by TOF-MS. In the S60 model, it was successfully traced back to an abnormal material crushing event in the crushing section. (Characteristic peak extraction results showed that high abundance isotope peaks of 2,2',4,4'-tetrabromodiphenyl ether (BDE-47) and decabromodiphenyl ether (BDE-209) were clearly detected in the sample, confirming the illegal mixing of high-risk flame-retardant plastics.) When sampling started at T=7s, the system precisely locked the sampling flow rate at 1.5L / min through the mass flow controller to ensure that the cutting particle size of the virtual impactor remained stable at 2.5 micrometers. At T=607s (i.e., the thermal desorption analysis stage after sampling), the sampling pipeline automatically performed a reverse high-temperature nitrogen purging (320℃) for 2 minutes to remove residues from the pipe wall and prevent memory effects from interfering with the next sampling. The system then linked the workshop fresh air system for power compensation.

[0077] Experimental verification and effect comparison:

[0078] To verify the effectiveness of the present invention, it was compared with the traditional timed sampling + offline GC-MS analysis scheme for a week.

[0079] Experimental setup:

[0080] Experimental group (this scheme): Intelligent trigger sampling is used, combined with non-negative constraint autoencoder (NNAE) for source parsing.

[0081] Control group (traditional scheme): Fixed-point sampling was performed every 2 hours, combined with positive definite matrix factorization (PMF) model for source analysis.

[0082] Analysis of experimental results: Comparison of pollution incident capture rates: During the experiment, a total of 12 short-term high-concentration pollution emission events occurred in the workshop (the true values ​​were determined using continuous CEMS data throughout the process).

[0083] Results: Due to the fixed sampling time, the control group only encountered two events by chance, with a capture rate of only 16.7%. The present invention successfully triggered and sampled 11 events, with a capture rate of 91.7%. The one event that was missed was because the event lasted only 3 seconds and was judged as interference noise by the system and filtered out. This actually proves the effectiveness of the anti-jitter logic.

[0084] Source resolution accuracy comparison: The correlation analysis was performed between the identified pollution source contributions and the actual production logs in the workshop (such as machine start-up time and material input). Please refer to the appendix for the specific results. Figure 3 (Comparison chart of pollution source apportionment time series).

[0085] Appendix Figure 3 Detailed explanation: The horizontal axis in the graph represents the duration of the workshop production process (unit: minutes), and the vertical axis represents the relative intensity of the impact of a specific pollution source (taking the volatilization of brominated flame retardants as an example) on ambient air quality. The data has been normalized (0-100).

[0086] The thick gray solid line in the figure represents the true value of the actual working conditions, that is, the actual pollution emission situation in the workshop. It can be clearly seen from the figure that there are two obvious peaks in the gray solid line, which represent two sudden pollution emission events that occurred during the experiment (such as the moment of feeding or equipment failure). This is the standard answer for comparison and verification.

[0087] The black solid line with dots in the figure represents the calculation result of the present invention. It can be seen that this curve closely follows the gray solid line, and the peak position and height are basically the same, indicating that the present invention can accurately and in real time reflect the real pollution fluctuations without obvious lag.

[0088] The black dashed curve with a cross in the figure represents the analysis result of the traditional scheme (timed sampling). Due to its low sampling frequency and the algorithm's tendency to smooth out the data, the curve did not capture the two key peaks, and the overall trend appeared flat and lagging, failing to reflect the real process risks.

[0089] Quantitative indicators: Statistical results show that the Pearson correlation coefficient between the analysis results of the present invention and the actual working conditions reaches 0.89, which is significantly higher than the 0.45 of the control group, proving the advantages of the present invention in dynamic source analysis.

[0090] Algorithm convergence verification: For the convergence performance of the physical constraint-based deep neural network (NNAE) model proposed in Section S60, please refer to the appendix. Figure 4 (Comparison and verification of algorithm convergence).

[0091] Appendix Figure 4 Detailed explanation: The horizontal axis in the graph represents the number of iterations (Epochs) in the training of the neural network model. The more iterations, the longer the computation time. The vertical axis represents the loss function value (Loss), which is the error between the model's calculated result and the true value. The smaller the value, the more accurate the model.

[0092] The black dotted lines in the figure represent ordinary neural network algorithms without physical constraints (such as nonnegativity and sparsity). As can be observed, the curve decreases slowly and exhibits obvious wave-like oscillations in the later stages of iteration, indicating that the calculation process is unstable and it is difficult to obtain a unique solution.

[0093] The black solid line in the figure represents the physical constraint network algorithm proposed in this invention. The curve drops almost vertically in the early stage of training, indicating that the error decreases rapidly and quickly turns into a straight line, representing that the model has reached an extremely stable state in a very short time.

[0094] A pentagram is specially marked in the figure to indicate the fast convergence point of the present invention. The results show that the algorithm of the present invention has reached the optimal convergence state in about 40 iterations.

[0095] Results: After introducing physical constraints (nonnegativity and sparsity regularization), the model loss function rapidly decreased within the first 50 iterations and converged completely around 100 iterations, with the reconstruction error stabilizing at 10. −4 In this verification experiment, the NNAE algorithm model was actually deployed on an industrial-grade embedded gateway (4GB memory, 1.5GHz clock speed) based on the ARM Cortex-A72 architecture. Test data showed that the average time to complete a complete source resolution calculation (including the iterative process) was only 180 milliseconds, and the CPU utilization rate was less than 35%, which fully met the real-time requirements of microsecond-level response in the field. This verified the engineering usability in low computing power environment. This result strongly proves that the algorithm of this invention has low computational overhead and fast convergence speed, and is very suitable for deployment on edge computing gateways with limited computing power.

[0096] Comprehensive performance evaluation: Experiments show that by employing a strategy of monitoring during normal standby and sampling triggered by anomalies, this invention reduces the usage of consumables such as adsorption tubes and filter membranes by more than 70% while obtaining the same amount of effective data, significantly reducing maintenance costs while ensuring the validity and timeliness of the data.

Claims

1. A method for screening pollutants from the dismantling of electronic waste based on non-targeting technology, characterized in that, Includes the following steps: The Internet of Things sensing network module (200) monitors environmental parameters, meteorological parameters, location information, working condition information and material characteristics. When the value of the characteristic index exceeds the preset threshold, a start signal is sent to the integrated online sampling module (100). The integrated online sampling module (100) receives the start signal to start the sampling pump, uses the inertial separation structure to grade and intercept particulate matter to the filter membrane, and uses the adsorption medium to capture volatile organic compounds. After sampling, the integrated online sampling module (100) starts the thermal desorption program to heat the filter membrane and the adsorption medium to desorb and vaporize the pollutants, and then introduces the sample into the mobile high-resolution analysis module (300). The mobile high-resolution analysis module (300) uses a soft ionization source to acquire mass spectrometry data in full scan mode to form a raw data stream; The original data stream is processed by the intelligent identification and tracing module (400) to generate a feature vector, which is then compared with the database to determine the type of pollutant. The intelligent identification and traceability module (400) combines the identified pollutant types, pollution source fingerprint database and the operating condition information, and uses chemometrics algorithms to calculate the contribution rate of dismantling processes or materials, and generates an analysis report.

2. The method for screening pollutants from electronic waste dismantling based on non-targeting technology according to claim 1, characterized in that, The steps for monitoring environmental parameters and material characteristics specifically include: collecting monitoring data using the Internet of Things sensing network module (200) equipped with a particulate matter sensor based on the principle of light scattering, a photoionization detector, and a material fingerprint recognition unit; And calculate the real-time pollution load index through the edge computing gateway; The real-time pollution load index is calculated by multiplying the ratio of particulate matter mass concentration to a preset particulate matter concentration benchmark threshold, the ratio of volatile organic compound concentration to a preset volatile organic compound concentration benchmark threshold, and the material spectral feature matching degree by the corresponding dimensionless weighting coefficients and then summing them up.

3. The method for screening pollutants from electronic waste dismantling based on non-targeting technology according to claim 2, characterized in that, The triggering logic for sending the start signal to the integrated online sampling module (100) is as follows: The edge computing gateway determines that the real-time pollution load index is greater than the trigger threshold, and determines that the duration of the state in which the real-time pollution load index is greater than the trigger threshold exceeds the set minimum confirmation time.

4. The method for screening pollutants from electronic waste dismantling based on non-targeting technology according to claim 1, characterized in that, The integrated online sampling module (100) controls the average airflow velocity at the nozzle outlet by adjusting the pumping flow rate of the downstream sampling pump, so that coarse particles larger than the cutting particle size cannot rush into the receiving tube, and fine particles smaller than the cutting particle size are deflected by the main airflow and guided through the built-in filter membrane. Furthermore, in the step of capturing volatile organic compounds using an adsorption medium, the airflow passes through an adsorption medium located downstream of the filter membrane, and the adsorption medium is sequentially filled with a weakly polar adsorbent and a strongly polar adsorbent along the airflow direction.

5. The method for screening pollutants from electronic waste dismantling based on non-targeting technology according to claim 1, characterized in that, The thermal desorption process includes: driving the valve group to operate through the flow path control unit, and using inert gas to establish a purge circuit to remove residual moisture and oxygen in the adsorption medium; The heating unit executes a flash heating strategy to control the heating rate and increase the temperature. In the step of introducing the sample into the mobile high-resolution analysis module (300), the sample is transported through a transmission pipeline with an inert inner wall and wrapped with a heat tracing and insulation layer on the outside, and the pipe wall temperature is maintained at a constant high temperature by a PID controller.

6. The method for screening pollutants from electronic waste dismantling based on non-targeting technology according to claim 1, characterized in that, In the step of acquiring mass spectrometry data, the mass spectrometer built into the mobile high-resolution analysis module (300) is protected by a steel wire rope vibration isolation system. An internal standard compound with a known precise mass number is periodically injected through an independent reference channel, and the mass axis coefficient is calibrated in reverse based on the measured flight time of the internal standard compound.

7. The method for screening pollutants from electronic waste dismantling based on non-targeting technology according to claim 1, characterized in that, The steps for processing the original data stream specifically include: constructing a deep neural network model based on physical constraints; The deep neural network model includes an encoder and a decoder. The encoder compresses the normalized concentration matrix constructed based on the feature vector into the hidden layer. The activation function of the hidden layer is a modified linear unit. The decoder reconstructs the hidden layer features back to the original dimension, and the connection weights of the decoder correspond to the source component spectrum. When updating weights through backpropagation, the updated negative weights are reset to zero using the projection operator.

8. The method for screening pollutants from electronic waste dismantling based on non-targeting technology according to claim 7, characterized in that, The training objective of the deep neural network model is achieved by minimizing the loss function; The loss function consists of a weighted reconstruction error term and a sparse regularization term; The weighted reconstruction error term includes the squared difference between the measured concentration and the network reconstruction output value, and is weighted using measurement uncertainty; The sparse regularization term comprises the product of the L1 regularization coefficient and the source contribution term corresponding to the hidden layer feature.

9. A method for screening pollutants from electronic waste dismantling based on non-targeting technology according to claim 8, characterized in that, Before using chemometric algorithms to calculate the contribution rate of different dismantling processes or materials, a data cleaning step is also included: Based on the timestamp of the mass spectrometry data, the meteorological and location data contained in the environmental parameters are upsampled and reconstructed, and data during calm periods with wind speeds less than a set value are identified and removed. The concentration matrix is ​​pre-analyzed using singular value decomposition, and the number of potential pollution sources is determined based on the cumulative contribution rate of eigenvalues.

10. The method for screening pollutants from electronic waste dismantling based on non-targeting technology according to claim 7, characterized in that, After analyzing the time contribution sequence of each pollution source, the conditional bivariate probability function is calculated by combining the synchronous wind field data contained in the environmental parameters, and the contribution in the time dimension is mapped to the orientation in the spatial dimension. When the total number of samples in a certain wind direction sector is less than the minimum statistical threshold, the probability value in the wind direction sector is forced to be zero, and the azimuth probability radar map of each pollution source is finally output.

Images