Method and system for analyzing health state of equipment under online monitoring of nano-sized particles

Abstract

The application discloses a method and system for analyzing the health state of equipment under online monitoring of nanoscale particles, and relates to the technical field of energy storage health monitoring. The method comprises the following steps: isokinetic sampling is performed on an energy storage battery module to obtain a nanoscale particle sample data set; a particle grading module is constructed to perform particle size screening, determine target particle data, and focus on constructing a single particle sequence flow; multi-dimensional feature time sequence data are constructed through synchronous monitoring of a double-angle light scattering detection area and a double-pulse laser-induced breakdown spectroscopy detection area; a charge-discharge cycle parameter is introduced to construct a multi-modal input tensor for health evaluation, and a device health state report is generated; operation and maintenance analysis is performed to generate a hierarchical operation and maintenance instruction. The method solves the technical problems of nanoscale sample sampling distortion, incomplete single-dimensional detection information, missing time sequence working condition coupling and disconnection of evaluation and operation and maintenance in the prior art, and achieves the technical effects of improving sampling fidelity, multi-modal feature fusion capability, dynamic health evaluation accuracy and operation and maintenance decision closed-loop.

Description

Technical Field

[0001] This invention relates to the field of energy storage health monitoring technology, specifically to a method and system for analyzing the health status of equipment under online monitoring of nanoscale microparticles. Background Technology

[0002] During long-term charge-discharge cycles, battery modules gradually experience electrode material degradation, electrolyte decomposition, and separator failure due to internal electrochemical side reactions, mechanical stress, and heat accumulation. The most critical precursor is the generation and release of nanoscale particles inside the battery, such as metal debris, carbon soot, lithium salt crystals, and oxide particles. Traditional battery management systems mainly rely on macroscopic electrothermal parameters such as voltage, current, and temperature for threshold alarms. However, the response to macroscopic parameters is lagging, often only showing significant fluctuations in the irreversible stage of thermal runaway, and they cannot reveal early kinetic information about the microscale physicochemical changes inside the battery. Current health status assessments for energy storage batteries struggle to achieve non-destructive extraction under constant velocity and kinetic conditions, leading to the loss of crucial physical evidence. Relying on optical scattering counting or spectral elemental analysis fails to simultaneously obtain the physical particle size distribution and chemical elemental composition of particulate matter, resulting in insufficient ability to trace the source of particulate matter. Furthermore, the inability to synchronize particulate matter emission characteristics with real-time charge and discharge conditions creates fragmented "data silos," hindering the construction of effective health degradation trend models. In addition, the health status output lacks a risk-level-based hierarchical operation and maintenance strategy generation mechanism, failing to guide on-site personnel in implementing differentiated response measures.

[0003] Therefore, current technologies suffer from technical problems such as nanoscale sample sampling distortion, incomplete single-dimensional detection information, lack of temporal condition coupling, and disconnect between evaluation and maintenance. Summary of the Invention

[0004] This application provides a method and system for analyzing the health status of equipment under online monitoring of nanoscale microparticles. It solves the technical problems existing in the prior art, such as nanoscale sample sampling distortion, incomplete single-dimensional detection information, lack of temporal condition coupling, and disconnect between assessment and operation and maintenance. It achieves the technical effects of improving sampling fidelity, multimodal feature fusion capability, accuracy of dynamic health assessment, and closed-loop operation and maintenance decision-making.

[0005] This application provides a method for analyzing the health status of equipment under online monitoring of nanoscale microparticles. The method includes: performing isokinetic sampling on an energy storage battery module using an isokinetic sampling probe to obtain a nanoscale microparticle sample dataset; constructing a microparticle classification module, using the microparticle classification module to traverse the nanoscale microparticle sample dataset for particle size screening, focusing on target microparticle data, and constructing a single-particle sequence stream; driving the single-particle sequence stream for online synchronous monitoring through a dual-angle light scattering detection area and a dual-pulse laser-induced breakdown spectroscopy detection area to construct multidimensional feature time-series data; introducing charge-discharge cycle parameters, aligning the multidimensional feature time-series data according to the charge-discharge cycle parameters, constructing a multimodal input tensor for health assessment, and generating an equipment health status report, the equipment health status report including a battery health status index and a thermal runaway risk level; and performing operation and maintenance analysis back to the energy storage battery module based on the battery health status index and the thermal runaway risk level to generate graded operation and maintenance instructions.

[0006] This application also provides a device health status analysis system under online monitoring of nanoscale microparticles. The system includes: a sample data sampling module for isokinetic sampling of the energy storage battery module using an isokinetic sampling probe to obtain a nanoscale microparticle sample dataset; a single particle sequence stream construction module for constructing a microparticle classification module, which traverses the nanoscale microparticle sample dataset through the microparticle classification module to perform particle size screening, determine target microparticle data for focusing, and construct a single particle sequence stream; a feature time-series data acquisition module for driving the single particle sequence stream for online synchronous monitoring through a dual-angle light scattering detection area and a dual-pulse laser-induced breakdown spectroscopy detection area to construct multidimensional feature time-series data; a health status report generation module for introducing charge-discharge cycle parameters, aligning the multidimensional feature time-series data according to the charge-discharge cycle parameters, constructing a multimodal input tensor for health assessment, and generating a device health status report, which includes a battery health status index and a thermal runaway risk level; and a graded operation and maintenance instruction generation module for backtracking to the energy storage battery module for operation and maintenance analysis based on the battery health status index and the thermal runaway risk level to generate graded operation and maintenance instructions.

[0007] This application proposes a method and system for analyzing the health status of equipment under online monitoring of nanoscale microparticles. This method involves isokinetic sampling of energy storage battery modules to obtain a dataset of nanoscale microparticle samples; constructing a microparticle grading module for particle size screening to determine target microparticle data and focusing on building a single-particle sequence stream; simultaneously monitoring through a dual-angle light scattering detection area and a dual-pulse laser-induced breakdown spectroscopy detection area to construct multi-dimensional characteristic time-series data; introducing charge-discharge cycle parameters to construct a multi-modal input tensor for health assessment and generating an equipment health status report; and performing operation and maintenance analysis to generate tiered operation and maintenance instructions. This approach solves the technical problems of nanoscale sample sampling distortion, incomplete single-dimensional detection information, missing time-series operating condition coupling, and disconnect between assessment and operation and maintenance in existing technologies. It achieves the technical effects of improving sampling fidelity, multi-modal feature fusion capability, accuracy of dynamic health assessment, and closed-loop operation and maintenance decision-making. Attached Figure Description

[0008] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the accompanying drawings of the embodiments of this disclosure will be briefly described below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0009] Figure 1 This is a schematic flowchart of a method for analyzing the health status of equipment under online monitoring of nanoscale microparticles, provided in an embodiment of this application.

[0010] Figure 2 The coarse particle size distribution is shown in the method for analyzing the health status of equipment under online monitoring of nanoscale microparticles provided in the embodiments of this application.

[0011] Figure 3 This is a schematic diagram of the structure of the device health status analysis system under online monitoring of nanoscale microparticles provided in an embodiment of this application.

[0012] Figure labeling: Sample data sampling module 10, single-particle sequence stream construction module 20, feature time series data acquisition module 30, health status report generation module 40, hierarchical operation and maintenance instruction generation module 50. Detailed Implementation

[0013] To further illustrate the technical means and effects adopted by the present invention in order to achieve the intended purpose, the following detailed description is provided in conjunction with the accompanying drawings and preferred embodiments, based on the specific implementation methods, structures, features and effects of the present invention.

[0014] This application provides a method for analyzing the health status of equipment under online monitoring of nanoscale microparticles, such as... Figure 1 As shown, the method includes: Step S100: The energy storage battery module is sampled at an constant velocity using a constant velocity sampling probe to obtain a nanoscale particle sample dataset.

[0015] Step S100 further includes that the isokinetic sampling probe comprises multiple sampling nozzles distributed along the internal gas flow channels of the energy storage battery module, each sampling nozzle is embedded with a hot-wire microflow sensor, and each sampling nozzle is connected to a micro air pump of the energy storage battery module through an independent pipeline to perform multi-channel synchronous isokinetic sampling of the energy storage battery module.

[0016] Preferably, the sampling nozzle is installed on the hollow tubular collection port on the wall of the gas flow channel inside the energy storage battery module, with its opening facing the airflow direction. It is used to extract the mixed gas flowing through this location, such as nano-sized particles. No fewer than two sampling nozzles are installed at different gas flow channel locations or different cross-sections of the same flow channel inside the same battery module. The number is determined according to the internal flow channel layout of the module, usually 3 to 8. The installation position of the sampling nozzle follows the direction of the preset cooling / exhaust airflow channel inside the battery module to ensure coverage of multiple characteristic cross-sections of the airflow from the inlet to the outlet, including but not limited to the gap channel between battery cells, the airflow gathering area above the pole, and the front section of the exhaust port at the end of the module. Hot-wire microflow sensors are fixedly installed on the inner wall or central support structure of the sampling nozzle. The sensing element is located inside the airflow channel between the inlet and outlet of the sampling nozzle, and is in direct contact with the gas flowing through the sampling nozzle. Hot-wire microflow sensors are local flow velocity measuring elements that work based on the principle of thermal equilibrium. The core of the sensor is a thin metal wire, such as a platinum wire or a tungsten wire, which is heated by electricity. The diameter is usually 5~25μm. The gas molecules flowing through the hot wire carry away heat, causing the resistance value of the hot wire to change. By measuring the change in resistance, the instantaneous local airflow velocity at the location of the sampling nozzle can be calculated.

[0017] Preferably, each sampling nozzle is connected to an independent gas delivery pipeline at its rear end. These pipelines are physically isolated, with no shared branches or junctions, ensuring independent transmission of gas samples collected by each nozzle and preventing sample mixing or cross-contamination across channels. The miniature suction pump is a small diaphragm or piston-type negative pressure pump installed outside the battery module or on its housing. Its intake port is connected to the outlet of all independent pipelines via a multi-port connector or manifold, providing stable negative pressure suction power for the entire sampling system. One end of each independent pipeline connects to the sampling nozzle outlet, and the other end connects to the inlet of the miniature suction pump or a manifold, forming a complete gas flow channel from the sampling nozzle → independent pipeline → suction pump. Multiple sampling nozzles operate simultaneously, forming multiple independent sampling channels for multi-channel synchronous isokinetic sampling of the energy storage battery module. Isokinetic sampling includes velocity matching conditions, isokinetic conditions, and continuous conditions throughout the sampling process, with the average airflow velocity at the sampling nozzle inlet cross-section being the key factor. The mainstream average velocity of the gas flow channel with respect to the characteristic cross section where the sampling nozzle is located The ratio satisfies , The allowable deviation limit is set according to the control accuracy of the sampling system, typically ±5% to ±10%; the airflow streamline entering the sampling nozzle is parallel and continuous with the undisturbed incoming flow streamline, and no streamline bending or acceleration effect is generated at the sampling nozzle inlet, thereby ensuring that the nanoscale particles suspended in the airflow enter the sampling nozzle with their original trajectory and original velocity in the flow channel, without particle deviation or enrichment due to inertial differences; the constant velocity state remains stable throughout the entire sampling duration, and the flow velocity fluctuation amplitude at the start and end of the sampling does not exceed the preset deviation threshold, ensuring the temporal continuity and state consistency of the sampling process.

[0018] Preferably, a hot-wire microfluidic sensor measures the actual suction velocity at the sampling nozzle inlet in real time and feeds the signal back to the speed control module of the micro-pump. The pump dynamically adjusts its output power based on the lowest velocity value or weighted average value fed back from each channel, changing the negative pressure at the inlet of each sampling nozzle until the inlet flow velocity of all sampling nozzles matches the inflow velocity at their respective positions. Finally, a nanoscale particle sample dataset is obtained. The suspended particles include solid or liquid particles released by electrochemical reactions inside the battery, electrode material stripping, or electrolyte decomposition. The particle size ranges from 10 nm to 1000 nm. The data recording content includes the particle information collected at each sampling moment, including but not limited to the sample collection timestamp, sampling point location identifier, sampling operating parameters, and the original particle concentration count.

[0019] Step S200: Construct a particle classification module. Through the particle classification module, traverse the nanoscale particle sample dataset to perform particle size screening, determine the target particle data for focusing, and construct a single particle sequence stream.

[0020] Step S200 further includes: performing coarse screening based on the nanoscale particle sample dataset to obtain a coarse-screened sample dataset; integrating a corona charging unit and a differential electromobility analyzer to construct a particle classification module; locating a buffer charging region, introducing the coarse-screened sample dataset into the buffer charging region for polarization processing, synchronizing the processing result to the corona charging unit for unipolar corona charging to generate charged particle parameters; synchronizing the charged particle parameters to the differential electromobility analyzer for progressive scanning to determine target particle data.

[0021] Preferably, an inertial impactor or cyclone separator is used to perform particle size pre-classification on the mixed aerosol corresponding to the nanoparticle sample dataset. This includes delivering the mixed aerosol output from the isokinetic sampling probe to the impact plate surface of the inertial impactor. The airflow is deflected at 90° at the nozzle outlet of the impactor. Larger particles, due to their greater inertial momentum, cannot follow the airflow and impact and adhere to the impact plate surface, thus being trapped. Smaller nanoparticles, due to their insufficient inertial momentum, follow the airflow deflection and pass through the air outlet behind the impact plate. This removes coarse particles larger than 1 μm in the sample, such as dust, fiber debris, and other non-battery-origin impurities, while retaining nanoparticles smaller than 1 μm, thus obtaining a coarse-screened sample dataset.

[0022] Preferably, the corona charging unit consists of a corona discharge needle electrode (made of tungsten or stainless steel, with a tip curvature radius of less than 100 μm) and a counter electrode ring. A non-uniform strong electric field is formed between the two by applying a DC high voltage. When the electric field strength near the needle tip exceeds the gas breakdown threshold, the gas molecules are ionized, generating a large number of free electrons and positive ions. The differential electromobility analyzer consists of a coaxial double-layer cylindrical electrode, including an inner electrode and an outer electrode, with a linearly adjustable DC voltage applied between the two electrodes. The gas outlet of the corona charging unit is connected to the aerosol inlet of the differential electromobility analyzer through a sealed pipeline, allowing the charged aerosol to directly enter the electric field classification zone of the differential electromobility analyzer, forming a continuous processing link from "charged" to "sorted". This determines the particle classification module, which is used to unipolarly charge the input aerosol particles and achieve continuous separation and sieving of particles of different sizes based on the differences in the mobility of the particles in the electric field.

[0023] Preferably, the buffer charging zone is a sealed pipe section, 10cm to 30cm in length, located between the outlet of the coarse sieving device and the inlet of the corona charging unit. It contains no electrodes or obstructions and provides a stable, low-turbulence transition space for the coarsely sieved aerosol samples. This allows the particles to achieve a stable, uniformly distributed flow state before entering the strong electric field charging zone. The coarsely sieved sample dataset is then introduced into the buffer charging zone for polarization treatment. Specifically, the particles in the coarsely sieved sample dataset are in a zero or weak electric field environment, where gas molecules randomly collide with the particle surface, causing the particle surface charge state to tend towards a steady-state distribution under thermodynamic equilibrium. This eliminates the static charge generated by frictional contact during coarse sieving and pipeline transport. The surface is non-uniformly charged; the uniformly suspended aerosol, after being polarized, flows out of the buffer charging region outlet and directly enters the corona discharge region of the corona charging unit. Under the strong electric field between the corona discharge needle and the counter electrode, the free electrons generated by ionization attach to the surface of the nanoscale particles passing through this region, giving the particles a net negative or net positive charge. The charge of the particles is proportional to the square of the particle size. When the particle size is less than 50 nm, the diffusion charging mechanism is dominant, and when the particle size is greater than 100 nm, the field-induced charging mechanism is dominant. This generates charged particle parameters, namely the charge carried by each nanoscale particle, the instantaneous electromobility of the particle, and the initial estimated value of the corresponding particle dynamic particle size.

[0024] Furthermore, step S200 also includes: synchronizing the charged particle parameters to the differential electromobility analyzer for rapid scanning to obtain particle size distribution overview data; traversing the particle size distribution overview data for concentration identification to determine key particle size ranges; driving the differential electromobility analyzer to perform fine scanning of the charged particle parameters according to the key particle size ranges to determine nanoparticle data; performing continuity verification based on the nanoparticle data, calculating particle size deviation values, dynamically adjusting the differential electromobility analyzer based on the particle size deviation values, updating the differential electromobility analyzer based on the adjustment results for closed-loop scanning, and determining the target particle data.

[0025] Preferably, the charged particle parameters are transported from the corona charging unit outlet to the aerosol inlet of the differential electromobility analyzer via an airtight delivery pipeline for progressive scanning. Specifically, the internal electrode voltage of the differential electromobility analyzer is scanned from the starting voltage to the ending voltage in a coarse-resolution scanning mode with a large voltage step (e.g., 10V~50V / step) and a short residence time (0.1s~0.3s per step). This allows for the acquisition of the overall profile of the particle size distribution in the aerosol sample within a short time, determining the macroscopic trend of particle concentration change with particle size, and obtaining general particle size distribution data, i.e., a coarse-grained particle size spectrum composed of particle concentration counts corresponding to each voltage step. This spectrum includes the horizontal axis scanning voltage value, the particle number concentration corresponding to each particle size channel on the vertical axis, the overall morphological characteristics of the spectrum (e.g., single-peak, double-peak, or multi-peak distribution), the approximate particle size range where the concentration peak is located, and the overall range of the particle size distribution. Figure 2 As shown.

[0026] Preferably, the particle size distribution overview data obtained by rapid scanning is numerically traversed, and the difference between the concentration value corresponding to each particle size channel and the concentration value of its adjacent channels is calculated. When the difference value changes from positive to negative, it is determined that there is a local concentration maximum at that position. When the difference value changes from negative to positive, it is determined that there is a local concentration minimum at that position. All identified local concentration maximums are traversed. For example, if the absolute value of the concentration corresponding to the local maximum exceeds 3 times the preset background noise concentration threshold, or if the proportion of the particle size channel corresponding to the peak value in the total number of particles exceeds 10% of the total concentration, then the top N peak values ​​(N is a preset positive integer, usually 1~3) are selected as the center position of the key particle size interval according to the concentration from high to low. The width of each key particle size interval is defined as the particle size range corresponding to the concentration on both sides of the peak value dropping to 50% of the peak concentration.

[0027] Preferably, the differential electromobility analyzer performs a fine scan of charged particle parameters within a key particle size range. The scan start voltage is set to the voltage value of the starting particle size corresponding to the key particle size range, and the scan end voltage is set to the voltage value of the ending particle size corresponding to the key particle size range. The voltage increment adopts a smaller voltage step (e.g., 1V~5V / step, corresponding to an equivalent particle size resolution of 0.5nm~2nm) and a longer residence time (1s~5s per step) to improve the particle size resolution and counting accuracy within this particle size range, and determine the nanoparticle data, including the precise particle size value, precise concentration value, fine particle size distribution spectrum, and characteristic particle size statistics such as peak particle size, median particle size, average particle size, and geometric standard deviation within the key particle size range.

[0028] Preferably, the particle size distribution curves in the nanoparticle data obtained by fine scanning are compared and analyzed with the theoretically expected distribution curves. Specifically, using the particle size-concentration data points obtained by fine scanning as discrete samples, a log-normal distribution curve is fitted using the least squares method. The residual between each measured data point and the theoretical concentration value at the corresponding particle size on the fitted curve is calculated. All residuals are squared, summed, and divided by the number of data points to obtain the root mean square residual as a goodness-of-fit index. When the goodness-of-fit index exceeds a preset threshold (usually 5%–10%), it is determined that there is an anomaly in data continuity. Then, the absolute difference between the peak particle size actually measured in each particle size channel during fine scanning and the peak particle size obtained in the previous scan is calculated using the following formula: ,in, The peak particle size was measured in this fine scan. For reference particle size values, when the particle size deviation exceeds the system's preset deviation tolerance (e.g., ±2nm), the scanning parameters of the differential electromobility analyzer are adjusted based on the sign (positive or negative) and magnitude of the deviation. For example, if the measured peak particle size is greater than the reference peak particle size, the starting and ending voltages of the differential electromobility analyzer are simultaneously reduced. If the deviation tends to increase, the dwell time per step is extended (to 2s–5s) to improve the signal-to-noise ratio and obtain more accurate counting statistics. Finally, the differential electromobility analyzer is updated based on the adjustment results to perform closed-loop scanning until the particle size deviation converges to the preset tolerance range (≤±1nm) and the rate of change of the deviation value for three consecutive scans is less than the preset rate of change threshold, or the number of iterations reaches the system's preset maximum number of closed-loop scans (e.g., 5 times). The target particle data is then determined, including the converged accurate peak particle size, accurate concentration value, particle size distribution width, and records of each adjustment parameter change during the closed-loop scanning process.

[0029] Preferably, the target particle data is focused, including spatial focusing and temporal focusing. Specifically, the aerosol gas flow containing particles corresponding to the target particle data is spatially compressed through a pneumatic focusing lens assembly. The pneumatic focusing lens assembly consists of multiple coaxially mounted annular orifice plates with progressively smaller apertures. When the aerosol flows through the focusing lens assembly, the gas carrier generates a pressure gradient change under the alternating action of the expansion and contraction chambers between the orifice plates. Due to inertia, the particles converge along the central axis, forming a collimated particle jet with a radial diffusion range ≤0.5mm centered on the axis. By adjusting the flow rate of the micro-pump and the back pressure at the rear outlet of the focusing lens assembly, the particles after spatial focusing tend to be uniformly distributed along the time axis, i.e., phase... The time interval between two adjacent particles passing through the detection area tends to be consistent. This is achieved by setting a damping orifice plate downstream of the focusing lens group. By changing the ventilation cross-sectional area of ​​the damping orifice plate, the airflow pulsation amplitude is adjusted to eliminate the instantaneous velocity fluctuations of particles caused by turbulence. This allows the particles to pass through the subsequent detection area in an orderly manner with an approximately constant linear velocity (deviation ≤ ±5%). Through the differential pressure diversion valve located at the outlet end of the focusing lens group, residual particles whose particle size is not within the target range are diverted and discharged with part of the carrier gas, while retaining particles within the particle size range corresponding to the target particle data. This ensures that the proportion of particles outside the target particle size range in the output particle group after focusing is less than 2% of the total number of particles. Finally, a single particle sequence flow, i.e., an ordered aerosol jet, is obtained.

[0030] In step S300, the single-particle sequence stream is driven to perform online synchronous monitoring through the dual-angle light scattering detection area and the dual-pulse laser-induced breakdown spectroscopy detection area to construct multi-dimensional characteristic time-series data.

[0031] Step S300 further includes: performing axial analysis based on the single-particle sequence flow to determine the axial direction; identifying the center based on the initial detection area of ​​dual-angle light scattering and the initial detection area of ​​dual-pulse laser-induced breakdown spectrum to determine the first detection center position and the second detection center position; aligning the first detection center position and the second detection center position with the axial direction to generate an alignment result; performing passage detection on the single-particle sequence according to the alignment result; adjusting the optical path of the initial detection area of ​​dual-angle light scattering according to the passage detection result to lock the dual-angle light scattering detection area; and adjusting the focus of the initial detection area of ​​dual-pulse laser-induced breakdown spectrum based on the dual-angle light scattering detection area to lock the dual-pulse laser-induced breakdown spectrum detection area.

[0032] Preferably, the dual-angle light scattering detection area and the dual-pulse laser-induced breakdown spectrum detection area are located. Specifically, by measuring the spatial distribution of particulate matter on the airflow cross section, the direction vector of the symmetry center line of the airflow in three-dimensional space is determined, and the axial direction is obtained. The dual-angle light scattering initial detection area refers to the default physical installation position of the light scattering detection component without any adjustment, including the intersection area of ​​the two laser beams in space. This area is the location of the detection spot in the initial state. The dual-pulse laser-induced breakdown spectrum initial detection area refers to the default physical installation position of the LIBS detection component without any adjustment, including the location of the focal point of the two pulse laser beams in space. To center the initial detection area of ​​dual-angle light scattering, a visible light guiding beam coaxial with the detection beam is projected onto the initial detection area. A calibration target is set at the optical path exit, and the two-dimensional coordinates of the center of the guiding beam spot on the calibration target are read and recorded as the first detection center position. To center the initial detection area of ​​dual-pulse LIBS, a continuous guiding beam coaxial with the pulsed laser is projected onto the initial detection area. A calibration target is set at the focal point, and the center coordinates of the narrowest point of the indicator spot are read and recorded as the second detection center position.

[0033] Preferably, through calculation and adjustment, the axis of the single-particle sequence flow is simultaneously made to pass through the first and second detection center positions, i.e., both detection centers are located on the same straight line and this straight line coincides with the axis direction; in the initial alignment state, the single-particle sequence flow is allowed to pass normally through the entire detection channel, and the dual-angle light scattering detection component is activated to perform actual particle detection, recording the particle count rate, the signal-to-noise ratio of the scattered light signal, and the rising edge width of the scattered light pulse to determine the passage detection result; then, based on the passage detection result, the optical path of the initial detection area of ​​the dual-angle light scattering is adjusted, including... The mounting base of the light scattering detection module is translated along the X and Y axes to move the first detection center position to the axis. If the translation of the first detection center position exceeds the adjustment range of the optical element, the angle of the reflector of the detection beam is adjusted to change the propagation path of the beam in space, so that the center of the detection spot is translated to the axis position. According to the signal-to-noise ratio data in the passage detection, the position of the beam expander or focusing lens of the detection beam is adjusted to change the diameter of the detection spot. Finally, the dual-angle light scattering detection area is locked so that the detection spot in the light scattering detection area remains in a constant position during equipment operation.

[0034] Preferably, the initial detection area of ​​the dual-pulse laser-induced breakdown spectrum is focused and adjusted according to the dual-angle light scattering detection area. Specifically, according to the system's preset dual detection area spacing value (e.g., 5cm to 20cm), the position at the distance from the first detection center is measured on the axis and set as the target position of the LIBS detection area. The focusing lens in the LIBS detection component is moved back and forth along the optical axis, so that the focusing beam waist position of the LIBS pulse laser is moved from the initial detection center position to the target position. After the first pulse laser is focused, the optical path folding mirror angle of the second pulse laser is adjusted so that the focal points of the two pulse lasers coincide in the same spatial coordinate, with the deviation controlled within ±0.05mm. Finally, the optical parameters after the focusing adjustment are recorded and fixed in the system configuration parameters, and the mechanical locking device of the corresponding adjustment mechanism is locked to keep the laser focal point of the LIBS detection area in a constant position during equipment operation, thus determining the locked dual-pulse laser-induced breakdown spectrum detection area.

[0035] Furthermore, step S300 also includes: setting a first scattered light collection optical path and a second scattered light collection optical path based on the dual-angle light scattering detection area; driving the single-particle sequence flow to perform synchronous monitoring through the first scattered light collection optical path and the second scattered light collection optical path to obtain forward scattered light parameters and side scattered light parameters; comparing the forward scattered light parameters and the side scattered light parameters to obtain single particle size parameters and single particle sphericity; extracting the distance value between the dual-angle light scattering detection area and the dual-pulse laser-induced breakdown spectrum detection area to calculate the particle flight velocity, and generating a delayed trigger signal based on the particle flight velocity; triggering a first pulse laser to ablate the single-particle sequence flow according to the delayed trigger signal to generate plasma data; triggering a second pulse laser to perform secondary heating processing on the plasma data based on a preset short interval to obtain enhanced atomic emission spectra; acquiring and analyzing the enhanced atomic emission spectra to obtain multi-element characteristic spectral line intensities; combining the multi-element characteristic spectral line intensities, the particle flight velocity, and the single-particle sequence flow according to time to construct the multi-dimensional characteristic time-series data.

[0036] Preferably, the first scattered light collection optical path refers to a forward scattered light collection optical system, whose optical axis coincides with the propagation direction of the probe beam, i.e., it is consistent with the direction of the single-particle sequence flow axis. This optical path includes a forward scattered light collection lens, a central light-shielding baffle, a forward scattered light focusing mirror, and a first photodetector. The second scattered light collection optical path refers to a side scattered light collection optical system, whose optical axis is orthogonal to the propagation direction of the probe beam at 90°, i.e., perpendicular to the single-particle sequence flow axis. This optical path includes a side scattered light collection lens group, a side scattered light focusing mirror, and a second photodetector. The centerlines of the two collection optical paths intersect at the center point of the probe spot in the dual-angle light scattering detection area, and the angle between them is 90°, forming an orthogonal dual-angle scattering signal acquisition architecture.

[0037] Preferably, a single particle in the single-particle sequence stream enters the detection spot area along the axis. The detection beam irradiates the particle surface and causes scattering. The first photodetector and the second photodetector start collecting data simultaneously at the same sampling rate (greater than or equal to 10 MHz, i.e., 10 data points are collected per microsecond) to obtain forward scattering parameters and side scattering parameters. The forward scattering parameters refer to the digital signal sequence obtained after amplification and analog-to-digital conversion of the time-domain analog signal output by the first photodetector, including the peak amplitude of the scattering pulse, the integral area of ​​the scattering pulse, the width of the scattering pulse, and the symmetry of the scattering pulse waveform. The side scattering parameters refer to the time-domain digital signal sequence output by the second photodetector, including the peak amplitude of the side scattering pulse, the integral area of ​​the side scattering pulse, the width of the side scattering pulse, and the symmetry of the side scattering pulse waveform. Based on Mie scattering theory, the intensity of forward-scattered light from a particle at the probe beam wavelength is positively correlated with the square of its particle size. The specific calculation process includes extracting the pulse peak amplitude of the forward-scattered light parameters and the pulse peak amplitude of the side-scattered light parameters; calculating the ratio of forward scattering to side scattering; using a pre-calibrated particle size-scattered light intensity standard curve, mapping the forward scattering peak amplitude to the equivalent sphere diameter; and outputting the particle size parameter of a single particle. Sphericity is defined as the degree to which the actual shape of the particle closely approximates an ideal sphere. It is calculated based on the difference in the intensity distribution of forward and side scattering light. Specifically, the ratio of the integrated area of ​​the forward-scattered pulse to the integrated area of ​​the side-scattered pulse, and the ratio of the width of the forward-scattered pulse to the width of the side-scattered pulse are calculated. These two ratios are compared with the reference ratio of a standard spherical particle to calculate the deviation. The sphericity value ranges from 0 to 1.0, where 1.0 represents a perfect sphere. A lower value indicates a greater deviation from spherical shape. The calculation formula is as follows:

[0038] Where S represents the sphericity value, and α and β are calibration coefficients. , This is a reference ratio for standard spherical particles. The integral area of ​​the forward-scattering pulse. The integral area of ​​the side-scattered pulse. The forward scattering pulse width, The width of the lateral scattering pulse.

[0039] Preferably, the spacing value is the straight-line spatial distance between the center position of the dual-angle light scattering detection area and the center position of the dual-pulse LIBS detection area along the axis of the single-particle sequence flow. The particle flight velocity is indirectly calculated by the time difference of the forward-scattered light pulses. Specifically, in the dual-angle light scattering detection area, the peak time of the forward-scattered light pulse generated when each particle passes through the detection spot is continuously recorded to obtain a timestamp queue with the particle arrival time as the sequence. Since the spacing between the two detection areas is known and the single-particle sequence flow moves at an approximately uniform speed during transmission, the initial value of the particle flight velocity is obtained by calculating the carrier gas flow rate and flow cross-sectional area under pre-calibrated conditions. ,in, Let Q be the particle velocity, Q be the carrier gas volumetric flow rate, and A be the cross-sectional area of ​​the flow channel in the detection section. Then, the particle velocity is divided by the distance between the two detection zones to obtain the time required for the particle to travel from the center of the light scattering detection zone to the center of the LIBS detection zone, i.e., the delay time. This delay time is written into the configuration register of the digital delay pulse generator as a delay parameter. When the dual-angle light scattering detection zone detects a valid particle event, the pulse signal output by the photodetector is used as a start / stop signal input to the delay pulse generator. After receiving this trigger signal, the delay pulse generator outputs a TTL level pulse, i.e., the delay trigger signal.

[0040] Preferably, the first pulse laser is a Q-switched pulsed solid-state laser with a pulse width of 5ns to 20ns and a single pulse energy of 50mJ to 200mJ. When the delayed trigger signal reaches the Q-switch trigger port of the first pulse laser, the laser instantaneously releases its stored energy and emits a single high-energy laser pulse. This laser pulse is focused onto the LIBS detection area by a focusing lens. When the target particle in the single-particle sequence stream happens to fly to this focal point, the laser energy density reaches GW / cm². 2 On a nanosecond timescale, after particulate matter absorbs laser energy, its temperature rises sharply, and the matter vaporizes, atomizes, and ionizes, forming a plasma composed of free electrons, ions, and neutral atoms. The raw spectral data generated in this process is plasma data.

[0041] Preferably, the preset short interval is the time difference between the emission time of the first pulse laser and the emission time of the second pulse laser, with a value ranging from 0.5 μs to 2 μs. This time difference is less than the characteristic cooling time of the plasma. The second pulse laser and the first pulse laser originate from the same laser system, and their wavelength and pulse width are consistent with the first pulse. The single pulse energy is 50% to 80% of that of the first pulse. After the preset short interval timing ends, the second pulse laser is automatically emitted and focused at the same spatial position along the same optical path as the first pulse laser. The energy of the second pulse laser is injected into the existing plasma cloud, re-exciting and reheating the atoms and ions in the plasma, further increasing the excitation temperature of the plasma (by about 10% to 30%), and extending the luminescence duration of the plasma. The atomic spectral intensity emitted by the reheated plasma is 3 to 10 times higher than that of the single-pulse excitation spectral intensity. This spectrum is the enhanced atomic emission spectrum.

[0042] Preferably, enhanced atomic emission spectra are acquired using an echelle grating spectrometer. The acquisition trigger signal is synchronized with the Q-switching trigger signal of the second pulsed laser. The acquisition window width is 1μs~10μs, and the acquisition mode is gated integration mode, meaning the detector is turned off during non-acquisition periods to reduce dark noise. The enhanced atomic emission spectra are then analyzed. Specifically, the continuous radiation background portion of the spectrum is polynomial-fitted, and the fitted background is subtracted from the original spectrum. Local maxima exceeding a preset threshold (typically three times the root mean square noise of the background) are searched, and the wavelength position corresponding to each local maximum is recorded. The wavelength position of each spectral peak is matched with the elemental characteristic wavelengths in the standard atomic spectral database to determine the element type corresponding to the peak. For each identified elemental characteristic peak, its peak intensity value or peak integral area value is extracted to obtain multi-element characteristic spectral line intensities. Finally, the multi-element characteristic spectral line intensities, particle flight velocities, and single-particle sequence streams are arranged and combined chronologically to obtain structured multi-dimensional characteristic time-series data.

[0043] Step S400: Introduce charge-discharge cycle parameters, align the multidimensional feature time series data according to the charge-discharge cycle parameters, construct a multimodal input tensor for health assessment, and generate a device health status report. The device health status report includes a battery health status index and a thermal runaway risk level.

[0044] Step S400 further includes: traversing the multidimensional feature time-series data to extract multiple particle feature vectors; performing charge and discharge sampling on the energy storage battery pack at a fixed sampling frequency to obtain charge and discharge cycle parameters; resampling the multiple particle feature vectors according to the time series of the charge and discharge cycle parameters to construct a particle feature matrix; concatenating the particle feature matrix with the charge and discharge cycle parameters according to feature dimensions to construct a multimodal input tensor; performing interactive analysis based on the multimodal input tensor using a graph attention network to calculate a battery health status index; performing degradation analysis based on the multimodal input tensor using a dilated causal convolutional network to calculate a thermal runaway risk level; and fusing the battery health status index with the thermal runaway risk level to construct the device health status report.

[0045] Preferably, the multi-dimensional feature time-series data is traversed, and each particulate record in the multi-dimensional feature time-series data matrix is ​​read row by row to determine multiple particle feature vectors, including particle size parameters, sphericity, intensity of characteristic spectral lines of each element, and particle flight velocity. The fixed sampling frequency refers to the reciprocal of the time interval for the energy storage battery management system to collect data on the battery pack operating parameters, usually set to 1Hz to 10Hz. The energy storage battery pack is charged and discharged through voltage sensors, current sensors, temperature sensors, and energy accumulators to obtain charge and discharge cycle parameters, that is, a complete set of parameters reflecting the battery's charge and discharge working state collected at each fixed sampling time, including at least the battery pack terminal voltage, charge and discharge current, voltage range of each cell, battery pack maximum temperature, minimum temperature, average temperature, temperature rise rate, charge and discharge capacity, accumulated charge and discharge energy, current cycle number, current operating state, and current state of charge (SOC).

[0046] Furthermore, step S400 also includes: analyzing the charge-discharge state changes of the plurality of particle feature vectors according to the time series of the charge-discharge cycle parameters to identify multiple battery operating stages, including a charging stage, a discharging stage, and a resting stage; resampling based on the charging stage and the plurality of particle feature vectors to obtain charging particle emission event density data; resampling based on the discharging stage and the plurality of particle feature vectors to obtain discharging particle emission event density data; resampling based on the resting stage and the plurality of particle feature vectors to obtain resting particle emission event density data; performing weighted interpolation based on the charging particle emission event density data, the discharging particle emission event density data, and the resting particle emission event density data to generate particle feature estimates; and synchronizing the particle feature estimates over time to construct the particle feature matrix.

[0047] Preferably, multiple particle feature vectors are resampled according to the time series of charge-discharge cycle parameters. Specifically, the current value in the charge-discharge cycle parameters is read point by point in time order. For each sampling time point, its operating stage category is determined according to the current value at that time, and a status label is assigned. Adjacent sampling points with the same status label in the time series are merged into a continuous operating stage segment. Each segment is uniquely defined by its start time, end time, and stage type (charging / discharging / resting). All identified continuous stage segments are organized chronologically, identifying multiple battery operation stages, including charging, discharging, and resting stages. Then, all charging stage segments are used as resampling time windows, and the time range covered by each charging stage segment is used as an independent charging sub-window. All particle feature vectors are traversed, and all particle feature vectors whose timestamps fall within any charging stage segment are selected to form a subset of charging events. For each charging stage segment, the particle count density within that segment is calculated. Then, the arithmetic mean of all particle feature vectors within that segment is calculated by dimension to obtain the average particle feature vector of that segment. The emission event density value of each charging stage segment is bound to the average particle feature vector of that segment to form the comprehensive resampling result of that segment, thus determining the charging particle emission event density data.

[0048] Preferably, consistent with the calculation logic of resampling during the charging phase, the particle feature vectors whose timestamps fall into any discharge phase segment are selected to form a subset of discharge events. For each discharge phase segment, the discharge particle emission event density of that segment is calculated, and the arithmetic mean of each dimension of all particle feature vectors within that discharge phase segment is calculated to obtain the average particle feature vector of that segment. The emission event density value of the discharge phase segment is bound to the average particle feature vector of that segment, and the resampling results of all discharge phase segments are arranged in chronological order to form discharge particle emission event density data. Similarly, consistent with the calculation logic of resampling during the charging and discharging phases, the particle feature vectors whose timestamps fall into any resting phase segment are selected to form a subset of resting events. For each resting phase segment, the resting particle emission event density of that segment is calculated, and the arithmetic mean of each dimension of all particle feature vectors within that resting phase segment is calculated. The emission event density value of the resting phase segment is bound to the average particle feature vector of that segment, and the resampling results of all resting phase segments are arranged in chronological order to form resting particle emission event density data.

[0049] Preferably, the emission density data of charging particles, discharging particles, and stationary particles are merged in chronological order to form a reference point sequence with the start time of each segment as the time coordinate and the average particle feature vector of each segment as the value. The sampling time of the charge / discharge cycle parameters is used as the interpolation target position. At each target position, a particle feature vector estimate needs to be estimated. For each target time point, the K nearest reference points (K is set to 2~4) are taken before and after it, and the particle feature vector estimate at that time is calculated using the inverse distance weighting method. In addition to the distance weighting, a stage type matching weight is introduced. If the target time point is located within the charging stage segment, the weight from the charging stage reference point is multiplied by a coefficient greater than 1, such as 1.2~1.5. The weights from the discharging stage and stationary stage reference points are not multiplied by this coefficient or are multiplied by a coefficient less than 1 to obtain the particle feature estimate. Finally, the particle feature estimates are synchronized in time to construct a particle feature matrix. Each element in the matrix is ​​a statistical aggregate value of that feature dimension within the corresponding time interval.

[0050] Preferably, the particle feature matrix and the charge-discharge cycle parameters are concatenated according to the feature dimension. That is, each row of the new matrix is ​​formed by horizontally connecting the particle aggregate feature vector within the time interval and the charge-discharge cycle parameter vector at the corresponding moment in the time interval, thus constructing a multimodal input tensor, where the row direction is the time step dimension and the column direction is the feature dimension.

[0051] Furthermore, step S400 also includes: performing feature analysis on the multimodal input tensor according to time steps to extract particle features; using the particle features and the charge / discharge cycle parameters as graph nodes, connecting the graph nodes in time order to construct multiple directed edges; integrating the multiple directed edges to generate a dynamic time-series graph; aggregating node features through the graph attention network to construct a node feature sequence; performing adjacent attention analysis on the graph nodes to calculate multiple attention weight parameters; and performing a fully connected mapping based on the node feature sequence according to the multiple attention weight parameters to generate the battery health state index.

[0052] Preferably, for each time window in the multimodal input tensor, the particle features corresponding to that time step are extracted row by row, including particle features and charge / discharge cycle parameters. The particle features and charge / discharge cycle parameters are used as graph nodes, and the graph nodes are traversed and connected in chronological order to construct multiple directed edges. For each node, its outgoing edges point to all nodes that satisfy the step size condition after it in time order, and its incoming edges come from all nodes that satisfy the step size condition in time order before it. Then, the multiple directed edges are integrated to generate a dynamic time series graph. All edges in the graph point from nodes with smaller time indices to nodes with larger time indices. Each node is only connected to nodes within K time steps before and after it. The adjacency matrix is ​​a sparse matrix with bandwidth constraints. The graph attention network receives a dynamic temporal graph as input and performs feature aggregation and update for each node. Specifically, a trainable linear transformation matrix is ​​applied to the initial feature vector of each node to map it to a high-dimensional feature space. For each node, its neighborhood is defined as the set of all predecessor nodes with directed edges pointing to it. A multi-head attention mechanism is used for calculation. The attention coefficients are calculated for the transformed features of the current node and the transformed features of each predecessor node, respectively. The transformed features of all predecessor nodes are weighted and summed according to the normalized attention weights to obtain the aggregation result of the head, thus obtaining the final aggregated features of the node. All nodes are arranged in chronological order with their aggregated features to form a node feature sequence.

[0053] Preferably, adjacency attention analysis is performed by traversing the graph nodes. Specifically, for each directed edge in the edge set, the node aggregation features at both ends of the edge are extracted. The attention weight parameters corresponding to each directed edge are calculated to obtain the attention weight parameters of the edge, which characterize the influence strength between node pairs, and thus output multiple attention weight parameters. Then, the node feature sequence is aggregated using multiple attention weight parameters, including collecting the attention weight parameters of all incoming edges pointing to each node, applying the mean of the weight parameters of all incoming edges of the node as a weighting coefficient to the aggregated features of the node, and applying the weights of all incoming edges of the node to all nodes. The feature sequence is obtained by weighting the points; the weighted feature sequence is then subjected to global average pooling along the node dimension to compress the features of multiple nodes into a single feature vector, which is then input into a fully connected layer network. This fully connected layer contains multiple linear transformation layers, each consisting of matrix multiplication and a nonlinear activation function. The scalar output of the fully connected layer is mapped to the [0, 1] interval by the Sigmoid activation function to generate a battery health state index, which represents a quantitative estimate of the current battery capacity retention rate. Its value ranges from 0 to 1, where 1 represents a health state equivalent to a new battery and 0 represents a degradation state that has reached the end-of-life threshold.

[0054] Preferably, the dilated causal convolutional network is a convolutional neural network architecture designed for temporal data, followed by degradation analysis. Specifically, the network consists of multiple stacked dilated causal convolutional layers, each containing multiple convolutional kernels (size 2-4). The dilation rate of each layer increases exponentially with each subsequent layer. Each convolutional layer performs a sliding convolution operation on the input temporal data, outputting a new temporal feature map. The deeper the layers, the wider the receptive field coverage of the output sequence over a wider time range. Each layer's output is residually connected to the input. The output of the last dilated causal convolutional layer is then processed by global average pooling and a fully connected layer, mapped to a three-dimensional output vector, which is then normalized to three categories using a softmax activation function. The probability distribution corresponds to low-risk, medium-risk, and high-risk levels, respectively. In the low-risk level, the current battery pack shows no signs of abnormal degradation, and particulate emission characteristics and charge / discharge parameters are within normal fluctuation ranges. In the medium-risk level, continuous abnormal particulate emissions are detected and strongly correlated with specific charge / discharge stages, indicating localized degradation within the battery but not yet triggering a chain reaction. In the high-risk level, particulate emission concentration increases sharply, and multi-element spectral lines are simultaneously enhanced, accompanied by abnormal voltage fluctuations or excessive temperature rise rates in charge / discharge parameters, indicating that thermal runaway triggering conditions are approaching. The level corresponding to the maximum of the three probability values ​​is the thermal runaway risk level calculated for that time window. Finally, the battery health status index and thermal runaway risk level are integrated to construct an equipment health status report and output it to the energy storage system's operation and maintenance management platform or cloud monitoring system.

[0055] Step S500: Based on the battery health status index and the thermal runaway risk level, backtrack to the energy storage battery module for operation and maintenance analysis, and generate hierarchical operation and maintenance instructions.

[0056] Preferably, the battery health status index and thermal runaway risk level in the health status report are traced back to the energy storage battery module to determine the spatial location of the particulate sample data that generated the health status report, the charge / discharge cycle stage (charging / discharging / resting) covered by the health status report, and the corresponding absolute time window. This is used to locate the operating segment that causes the current health status degradation behavior, and to determine the main contributing feature dimensions that drive the decrease in SOH index and the increase in thermal runaway risk level, such as abnormal increase in the spectral intensity of specific elements, sudden increase in particulate matter concentration in a specific particle size range, and abnormal density of particulate emission events in a specific charge / discharge stage. Then, based on the traceback results and combined with a preset operation and maintenance judgment rule base, the current state of the energy storage battery module is used to diagnose operation and maintenance needs, and output graded operation and maintenance instructions, such as no operation instructions, continuous monitoring instructions, operating parameter adjustment instructions, planned maintenance instructions, and emergency safety isolation instructions, forming a complete management closed loop of "assessment → decision → execution → confirmation", thereby ensuring improved accuracy of dynamic health assessment and closed-loop operation and maintenance decision-making.

[0057] In the above text, refer to Figure 1This paper describes in detail a method for analyzing the health status of a device under online monitoring of nanoscale microparticles according to embodiments of the present invention. Next, reference will be made to... Figure 3 This invention describes a device health status analysis system based on online monitoring of nanoscale microparticles according to an embodiment of the present invention.

[0058] The equipment health status analysis system under online monitoring of nanoscale microparticles according to embodiments of the present invention is used to solve the technical problems existing in the prior art, such as nanoscale sample sampling distortion, incomplete single-dimensional detection information, lack of temporal condition coupling, and disconnect between assessment and maintenance. It achieves the technical effects of improving sampling fidelity, multimodal feature fusion capability, accuracy of dynamic health assessment, and closed-loop operation and maintenance decision-making. Figure 3 As shown, the equipment health status analysis system under online monitoring of nanoscale microparticles includes: a sample data sampling module 10, a single particle sequence stream construction module 20, a characteristic time series data acquisition module 30, a health status report generation module 40, and a hierarchical operation and maintenance instruction generation module 50.

[0059] The sample data sampling module 10 is used to perform isokinetic sampling on the energy storage battery module using an isokinetic sampling probe to obtain a nanoscale particle sample dataset. The single particle sequence stream construction module 20 is used to construct a particle classification module, which traverses the nanoscale particle sample dataset through the particle classification module to perform particle size screening, determine target particle data for focusing, and construct a single particle sequence stream. The feature time series data acquisition module 30 is used to drive the single particle sequence stream for online synchronous monitoring through a dual-angle light scattering detection area and a dual-pulse laser-induced breakdown spectroscopy detection area to construct multi-dimensional feature time series data. The health status report generation module 40 is used to introduce charge-discharge cycle parameters, align the multi-dimensional feature time series data according to the charge-discharge cycle parameters, construct a multi-modal input tensor for health assessment, and generate a device health status report, which includes a battery health status index and a thermal runaway risk level. The hierarchical operation and maintenance instruction generation module 50 is used to backtrack to the energy storage battery module for operation and maintenance analysis based on the battery health status index and the thermal runaway risk level, and generate hierarchical operation and maintenance instructions.

[0060] The specific configuration of the sample data sampling module 10 will be described in detail below. The sample data sampling module 10 further includes: the isokinetic sampling probe comprises multiple sampling nozzles distributed along the internal gas flow channels of the energy storage battery module, each sampling nozzle is embedded with a hot-wire microflow sensor, and each sampling nozzle is connected to a micro air pump of the energy storage battery module through an independent pipeline to perform multi-channel synchronous isokinetic sampling of the energy storage battery module.

[0061] The specific configuration of the single-particle sequence stream construction module 20 will be described in detail below. The single-particle sequence stream construction module 20 further includes: coarse screening based on the nanoscale particle sample dataset to obtain a coarse-screened sample dataset; integrating a corona charging unit and a differential electromobility analyzer to construct a particle classification module; locating a buffer charging region, introducing the coarse-screened sample dataset into the buffer charging region for polarization processing, synchronizing the processing result to the corona charging unit for unipolar corona charging to generate charged particle parameters; and synchronizing the charged particle parameters to the differential electromobility analyzer for progressive scanning to determine target particle data.

[0062] The specific configuration of the single-particle sequence stream construction module 20 will be described in detail below. The single-particle sequence stream construction module 20 further includes: synchronizing the charged particle parameters to the differential electromobility analyzer for rapid scanning to obtain particle size distribution overview data; traversing the particle size distribution overview data for concentration identification to determine key particle size ranges; driving the differential electromobility analyzer to perform fine scanning of the charged particle parameters according to the key particle size ranges to determine nanoparticle data; performing continuity verification based on the nanoparticle data, calculating particle size deviation values, dynamically adjusting the differential electromobility analyzer based on the particle size deviation values, updating the differential electromobility analyzer based on the adjustment results for closed-loop scanning, and determining the target particle data.

[0063] The specific configuration of the feature time-series data acquisition module 30 will be described in detail below. The feature time-series data acquisition module 30 further includes: performing axial analysis based on the single-particle sequence stream to determine the axial direction; identifying the center based on the initial detection area of ​​dual-angle light scattering and the initial detection area of ​​dual-pulse laser-induced breakdown spectrum to determine the first detection center position and the second detection center position; aligning the first detection center position and the second detection center position with the axial direction to generate an alignment result; performing passage detection on the single-particle sequence based on the alignment result; adjusting the optical path of the initial detection area of ​​dual-angle light scattering based on the passage detection result to lock the dual-angle light scattering detection area; and adjusting the focus of the initial detection area of ​​dual-pulse laser-induced breakdown spectrum based on the dual-angle light scattering detection area to lock the dual-pulse laser-induced breakdown spectrum detection area.

[0064] The specific configuration of the feature time series data acquisition module 30 will be described in detail below. The feature time-series data acquisition module 30 further includes: setting a first scattered light collection optical path and a second scattered light collection optical path based on the dual-angle light scattering detection area; driving the single-particle sequence flow to perform synchronous monitoring through the first scattered light collection optical path and the second scattered light collection optical path to obtain forward scattered light parameters and side scattered light parameters; comparing the forward scattered light parameters and the side scattered light parameters to obtain single particle size parameters and single particle sphericity; extracting the distance value between the dual-angle light scattering detection area and the dual-pulse laser-induced breakdown spectrum detection area to calculate the particle flight velocity, and generating a delayed trigger signal based on the particle flight velocity; triggering a first pulse laser to ablate the single-particle sequence flow according to the delayed trigger signal to generate plasma data, and triggering a second pulse laser to perform secondary heating processing on the plasma data based on a preset short interval to obtain enhanced atomic emission spectra; acquiring and analyzing the enhanced atomic emission spectra to obtain multi-element feature spectral line intensities; and combining the multi-element feature spectral line intensities, the particle flight velocity, and the single-particle sequence flow according to time intervals to construct the multi-dimensional feature time-series data.

[0065] The specific configuration of the health status report generation module 40 will be described in detail below. The health status report generation module 40 further includes: extracting multiple particle feature vectors by traversing the multidimensional feature time-series data; collecting charge and discharge data of the energy storage battery pack at a fixed sampling frequency to obtain charge and discharge cycle parameters; resampling the multiple particle feature vectors according to the time series of the charge and discharge cycle parameters to construct a particle feature matrix; concatenating the particle feature matrix and the charge and discharge cycle parameters according to feature dimensions to construct a multimodal input tensor; performing interactive analysis based on the multimodal input tensor using a graph attention network to calculate a battery health status index; performing degradation analysis based on the multimodal input tensor using a dilated causal convolutional network to calculate a thermal runaway risk level; and fusing the battery health status index and the thermal runaway risk level to construct the device health status report.

[0066] The specific configuration of the health status report generation module 40 will be described in detail below. The health status report generation module 40 further includes: analyzing the charge / discharge state changes of the multiple particle feature vectors according to the time series of the charge / discharge cycle parameters, identifying multiple battery operating stages, including a charging stage, a discharging stage, and a resting stage; resampling based on the charging stage and the multiple particle feature vectors to obtain charging particle emission event density data; resampling based on the discharging stage and the multiple particle feature vectors to obtain discharging particle emission event density data; resampling based on the resting stage and the multiple particle feature vectors to obtain resting particle emission event density data; performing weighted interpolation based on the charging particle emission event density data, the discharging particle emission event density data, and the resting particle emission event density data to generate particle feature estimates; and synchronizing the particle feature estimates over time to construct the particle feature matrix.

[0067] The specific configuration of the health status report generation module 40 will be described in detail below. The health status report generation module 40 further includes: performing feature analysis on the multimodal input tensor according to time steps to extract particle features; using the particle features and the charge / discharge cycle parameters as graph nodes, connecting the graph nodes in chronological order to construct multiple directed edges; integrating the multiple directed edges to generate a dynamic time-series graph; aggregating node features through the graph attention network to construct a node feature sequence; performing adjacent attention analysis on the graph nodes to calculate multiple attention weight parameters; and generating the battery health status index based on the node feature sequence and the multiple attention weight parameters through a fully connected mapping.

[0068] The equipment health status analysis system under online monitoring of nanoscale microparticles provided in this embodiment of the invention can execute the equipment health status analysis method under online monitoring of nanoscale microparticles provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method.

[0069] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for analyzing the health status of equipment under online monitoring of nanoscale microparticles, characterized in that, The method includes: A nanoscale particle sample dataset was obtained by performing isokinetic sampling on the energy storage battery module using an isokinetic sampling probe. A particle classification module is constructed. The particle classification module is used to traverse the nanoscale particle sample dataset to perform particle size screening, determine the target particle data for focusing, and construct a single particle sequence stream. The single-particle sequence stream is driven by a dual-angle light scattering detection area and a dual-pulse laser-induced breakdown spectroscopy detection area for online synchronous monitoring, thereby constructing multi-dimensional characteristic time-series data. By introducing charge-discharge cycle parameters, the multidimensional feature time-series data is time-aligned according to the charge-discharge cycle parameters, a multimodal input tensor is constructed for health assessment, and a device health status report is generated. The device health status report includes the battery health status index and the thermal runaway risk level. Based on the battery health status index and the thermal runaway risk level, the operation and maintenance analysis is traced back to the energy storage battery module to generate hierarchical operation and maintenance instructions. The method involves driving the single-particle sequence stream for online synchronous monitoring through a dual-angle light scattering detection region and a dual-pulse laser-induced breakdown spectroscopy detection region to construct multi-dimensional characteristic time-series data. The method includes: A first scattered light collection optical path and a second scattered light collection optical path are set based on the dual-angle light scattering detection area; The single-particle sequence stream is synchronously monitored by driving the first scattered light collection optical path and the second scattered light collection optical path to obtain forward scattered light parameters and side scattered light parameters. Based on the comparison of the forward scattered light parameters and the side scattered light parameters, the particle size parameters and sphericity of individual particles are obtained; The distance value between the dual-angle light scattering detection area and the dual-pulse laser-induced breakdown spectrum detection area is extracted to calculate the particle flight velocity, and a delayed trigger signal is generated based on the particle flight velocity. According to the delayed trigger signal, a first pulse laser is triggered to ablate the single-particle sequence flow to generate plasma data. Based on a preset short interval, a second pulse laser is triggered to perform secondary heating processing on the plasma data to obtain enhanced atomic emission spectra. The enhanced atomic emission spectra are collected and analyzed to obtain the intensity of multi-element characteristic spectral lines. The multi-element characteristic spectral line intensity, the particle flight velocity, and the single particle sequence stream are combined according to time to construct the multi-dimensional characteristic time-series data.

2. The method for analyzing the health status of equipment under online monitoring of nanoscale microparticles as described in claim 1, characterized in that, The isokinetic sampling probe includes multiple sampling nozzles distributed along the internal gas flow channels of the energy storage battery module. Each sampling nozzle is embedded with a hot-wire microflow sensor, and each sampling nozzle is connected to a micro air pump of the energy storage battery module through an independent pipeline to perform multi-channel synchronous isokinetic sampling of the energy storage battery module.

3. The method for analyzing the health status of equipment under online monitoring of nanoscale microparticles as described in claim 1, characterized in that, A particle classification module is constructed, and the particle size is screened by traversing the nanoscale particle sample dataset through the particle classification module to determine the target particle data. The method includes: Based on the nanoscale particle sample dataset, a coarse screening is performed to obtain a coarse screening sample dataset. A particle classification module is constructed by integrating a corona charging unit and a differential electromobility analyzer. The buffer charging region is located, and the coarse screening sample dataset is introduced into the buffer charging region for polarization processing. The processing results are synchronized to the corona charging unit for unipolar corona charging to generate charged particle parameters. The charged particle parameters are synchronized to the differential electromobility analyzer for progressive scanning to determine the target particle data.

4. The method for analyzing the health status of equipment under online monitoring of nanoscale microparticles as described in claim 3, characterized in that, The method involves synchronizing the charged particle parameters to the differential electromobility analyzer for progressive scanning to determine the target particle data, including: The charged particle parameters are synchronized to the differential electromobility analyzer for rapid scanning to obtain particle size distribution profile data; The concentration is identified by traversing the particle size distribution profile data to determine the key particle size range; The differential electromobility analyzer is driven according to the key particle size range to perform a fine scan of the charged particle parameters and determine the nanoparticle data; Based on the nanoparticle data, a continuity check is performed, the particle size deviation value is calculated, the differential electromobility analyzer is dynamically scanned and adjusted according to the particle size deviation value, and the differential electromobility analyzer is updated according to the adjustment result to perform closed-loop scanning to determine the target particle data.

5. The method for analyzing the health status of equipment under online monitoring of nanoscale microparticles as described in claim 1, characterized in that, The process and method for locating the dual-angle light scattering detection region and the dual-pulse laser-induced breakdown spectroscopy detection region include: Axial analysis is performed based on the single-particle sequence flow to determine the axis direction; The locations of the first and second detection centers are determined by center marking based on the initial detection areas of dual-angle light scattering and dual-pulse laser-induced breakdown spectrum. Align the first detection center position and the second detection center position with the axis direction to generate an alignment result; Based on the alignment result, the single particle sequence is subjected to passage detection, and based on the passage detection result, the optical path of the initial detection area of ​​dual-angle light scattering is adjusted to lock the dual-angle light scattering detection area; The initial detection area of ​​the dual-pulse laser-induced breakdown spectrum is focused and adjusted based on the dual-angle light scattering detection area to lock the dual-pulse laser-induced breakdown spectrum detection area.

6. The method for analyzing the health status of equipment under online monitoring of nanoscale microparticles as described in claim 1, characterized in that, The method involves aligning the multidimensional feature time-series data according to the charge-discharge cycle parameters, constructing a multimodal input tensor for health assessment, and generating a device health status report. Extract multiple particle feature vectors by traversing the multidimensional feature time-series data; The energy storage battery pack is charged and discharged at a fixed sampling frequency to obtain charge and discharge cycle parameters. The multiple particle feature vectors are resampled according to the time series of the charge-discharge cycle parameters to construct a particle feature matrix; The particle feature matrix and the charge / discharge cycle parameters are concatenated according to the feature dimension to construct a multimodal input tensor; Based on the multimodal input tensor, an interactive analysis is performed through a graph attention network to calculate the battery health status index; Degradation analysis is performed using a dilated causal convolutional network based on the multimodal input tensor to calculate the thermal runaway risk level; The battery health status index is fused with the thermal runaway risk level to construct the device health status report.

7. The method for analyzing the health status of equipment under online monitoring of nanoscale microparticles as described in claim 6, characterized in that, The method for resampling the multiple particle feature vectors according to the time series of the charge-discharge cycle parameters to construct a particle feature matrix includes: Based on the time series of the charge and discharge cycle parameters, the multiple particle feature vectors are analyzed to identify the charge and discharge state changes, and multiple battery operation stages are identified, including the charging stage, the discharging stage, and the resting stage. Based on the charging phase and combined with the multiple particle feature vectors, resampling is performed to obtain charging particle emission event density data; Based on the discharge stage and combined with the multiple particle feature vectors, resampling is performed to obtain discharge particle emission event density data; Based on the resting phase and combined with the multiple particle feature vectors, resampling is performed to obtain the resting particle emission event density data. Based on the charging particle emission event density data, the discharging particle emission event density data, and the stationary particle emission event density data, a weighted interpolation is performed to generate a particle feature estimate; The particle feature estimates are synchronized over time to construct the particle feature matrix.

8. The method for analyzing the health status of equipment under online monitoring of nanoscale microparticles as described in claim 6, characterized in that, Based on the multimodal input tensor, an interaction analysis is performed using a graph attention network to calculate the battery health status index. The method includes: The multimodal input tensor is subjected to feature analysis according to time steps to extract particle features; The particle features and the charge / discharge cycle parameters are used as graph nodes, and the graph nodes are connected in chronological order to construct multiple directed edges; The multiple directed edges are integrated to generate a dynamic time sequence graph; The node feature sequence is constructed by traversing the dynamic temporal graph through the graph attention network to aggregate node features. The graph nodes are traversed to perform adjacent attention analysis and calculate multiple attention weight parameters. The battery health status index is generated by performing a fully connected mapping based on the node feature sequence according to the multiple attention weight parameters.

9. A system for analyzing the health status of equipment under online monitoring of nanoscale microparticles, characterized in that, The system is used to implement the device health status analysis method under online monitoring of nanoscale microparticles as described in any one of claims 1 to 8, the system comprising: The sample data sampling module is used to perform isokinetic sampling on the energy storage battery module through an isokinetic sampling probe to obtain a nanoscale particle sample dataset. A single-particle sequence stream construction module is used to construct a particle classification module. The particle classification module traverses the nanoscale particle sample dataset to perform particle size screening, determines the target particle data for focusing, and constructs a single-particle sequence stream. The feature time series data acquisition module is used to drive the single particle sequence stream for online synchronous monitoring through the dual-angle light scattering detection area and the dual-pulse laser-induced breakdown spectroscopy detection area to construct multi-dimensional feature time series data. The health status report generation module is used to introduce charge-discharge cycle parameters, align the multi-dimensional feature time-series data according to the charge-discharge cycle parameters, construct a multi-modal input tensor for health assessment, and generate a device health status report. The device health status report includes a battery health status index and a thermal runaway risk level. The hierarchical operation and maintenance instruction generation module is used to perform operation and maintenance analysis on the energy storage battery module based on the battery health status index and the thermal runaway risk level, and generate hierarchical operation and maintenance instructions.

Images