An electrolyte fluid intelligent monitoring method and system based on multi-modal perception

Abstract

The application relates to an industrial fluid intelligent detection technology in the field of new-generation information technology, and discloses an electrolyte fluid intelligent monitoring method and system based on multi-modal perception, which has the technical scheme as follows: bottom layer signal synchronous acquisition and encapsulation, thermodynamic baseline compensation, orthogonal decoupling and dimension reduction pooling, time sequence high-order difference mapping, multi-node evidence fusion decision, comprehensive degradation evaluation and adaptive control; a multi-source physical sensing matrix excluding optical vision and a chain calculation architecture of soft and hard cooperation are adopted, and in sequence, thermodynamic baseline compensation, orthogonal feature decoupling, time sequence difference mapping and D-S evidence theory fusion are carried out, so that environmental drift interference and single-node false alarm can be effectively filtered out, early microcosmic degradation transients of the fluid can be accurately captured, and high-robustness industrial-grade adaptive early warning and chain protection are realized.

Description

Technical Field

[0001] This invention relates to intelligent detection technology for industrial fluids in the field of next-generation information technology, and more specifically, to an intelligent monitoring method and system for electrolyte fluids based on multimodal sensing. Background Technology

[0002] Traditional electrolyte fluid monitoring primarily relies on optical or visual imaging technologies, which are highly susceptible to failure due to light refraction or dissipation under complex industrial conditions such as turbid liquids, abundant microbubble formation, and light attenuation. Furthermore, existing single-mode sensors are easily affected by the combined interference of environmental thermodynamic drift and high-frequency hardware noise in real-world operating conditions, resulting in extremely high false alarm and false alarm rates for fluid monitoring systems. In addition, existing systems struggle to accurately detect early microscopic anomalies and deterioration accelerations within the fluid before a fault causes macroscopic limits to be exceeded. These shortcomings collectively lead to the technical problems of low accuracy, poor operational reliability, and the inability to achieve full lifecycle industrial-grade adaptive interlocking protection in existing monitoring technologies.

[0003] Therefore, the present invention provides an intelligent monitoring method and system for electrolyte fluid based on multimodal sensing, which improves the above-mentioned technical problems. Summary of the Invention

[0004] This disclosure aims to address the shortcomings of existing technologies by providing a method and system for intelligent monitoring of electrolyte fluids based on multimodal sensing. The invention employs a multi-source physical sensing matrix that excludes optical vision and a chain-like computing architecture that combines hardware and software. Through thermodynamic baseline compensation, orthogonal feature decoupling, temporal differential mapping, and DS evidence theory fusion, it effectively filters out environmental drift interference and single-node false alarms while accurately capturing early microscopic degradation transients of the fluid, thereby achieving highly robust industrial-grade adaptive early warning and cascading protection.

[0005] To achieve the above objectives, the present disclosure proposes the following technical solutions:

[0006] In a first aspect, embodiments of this disclosure propose an intelligent monitoring method for electrolyte fluid based on multimodal sensing, applied to a main control chip, comprising:

[0007] S1. Low-level signal synchronous acquisition and encapsulation: Drive the multi-modal physical sensing matrix to perform low-level signal synchronous acquisition, obtain and encapsulate a multi-source physical column vector sequence containing synchronization timestamps;

[0008] S2, Thermodynamic Baseline Compensation: Call the dynamic compensation algorithm module to perform thermodynamic baseline compensation on the multi-source physical column vector sequence, generate pure multimodal features and write them into the compensation feature area of ​​static random access memory (SRAM);

[0009] S3, Orthogonal Decoupling and Dimensional Reduction Pooling: Construct feature column vectors in the tensor space and perform orthogonal decoupling and dimensional reduction pooling to output pure integrated electrical feature components and independent acoustic scalar feature components, and push them into the first-in-first-out (FIFO) buffer queue;

[0010] S4, Temporal Higher-Order Differential Mapping: Call the digital differential filter to extract the feature component temporal data in the FIFO buffer queue according to pointer addressing, perform temporal higher-order differential mapping, and extract the absolute dynamic trend energy value;

[0011] S5. Multi-node evidence fusion decision: Substitute the absolute dynamic trend energy value and the corresponding decoupled scalar feature value into the dimensionless normalized exponential penalty model, calculate the basic probability distribution confidence, perform multi-node orthogonal fusion according to DS evidence theory, solve the final decision credibility probability and lock it to the TrustZone security isolation area in real time.

[0012] S6. Comprehensive Deterioration Assessment and Adaptive Control: The comprehensive degradation index of the electrolyte fluid is generated by engineering fusion mapping, and the comprehensive degradation index of the electrolyte fluid is compared with the first warning threshold and the second protection threshold for hardware branch determination, driving the external terminal to execute adaptive interlocking protection actions.

[0013] As a preferred embodiment of the present invention, step S2 involves thermodynamic baseline compensation of the multi-source physical column vector sequence, including:

[0014] Based on the real-time absolute temperature of the fluid, the preset standard reference absolute temperature, and the basic activation energy of the electrolyte, the original dynamic electrochemical impedance spectrum amplitude is subjected to exponential temperature compensation.

[0015] Based on the first-order change of the fluid's real-time absolute temperature relative to the standard reference absolute temperature and the first-order change of the fluid's real-time absolute pressure relative to the standard reference absolute pressure, and combined with the corresponding sensitivity compensation coefficient, the original ultrasonic attenuation coefficient is linearly compensated for temperature and pressure.

[0016] As a preferred embodiment of the present invention, step S3 involves performing orthogonal decoupling and dimensionality reduction pooling, including:

[0017] The pure multimodal features are concatenated into a multimodal feature column vector in contiguous memory address blocks. Calculate its autocorrelation similarity matrix by subtracting the identity matrix. Remove its own similarity components and perform row-by-row processing. Normalization generates cross-modal redundancy penalty orthogonal feature decoupling weight matrix ,satisfy:

[0018] ;

[0019] in, express The transpose of ; express The second norm; This indicates that normalization mapping is performed according to the matrix row dimensions; Represents the identity matrix of the corresponding dimension; Indicates the current sampling time;

[0020] Will Reverse action on the After stripping away the physically overlapping components, a decoupled fluid feature column vector is generated. , cut off the The former Each element serves as a high-dimensional electrical feature vector. and to Perform Euclidean norm pooling compression mapping to output pure synthesized electrical eigencomponents. ,satisfy:

[0021] ;

[0022] in, The corresponding number of discrete excitation frequency points; This represents the k-th component element of a high-dimensional electrical eigenvector; The last element is used as an independent acoustic scalar characteristic component. .

[0023] As a preferred embodiment of the present invention, step S4 involves extracting the absolute dynamic trend energy value, including:

[0024] The current scalar feature is read by the head pointer, and the historical feature of the corresponding period is retrieved by traversing the tail pointer according to the address offset.

[0025] The time series higher-order difference mapping is executed by calling the difference operator respectively. The first-order trend energy is determined based on the product of the first-order time difference gain coefficient and the absolute value of the first-order characteristic rate of change. The second-order trend energy is determined based on the product of the second-order time difference gain coefficient and the absolute value of the second-order characteristic rate of change.

[0026] The first-order trend energy and the second-order trend energy are weighted and summed to obtain the corresponding absolute dynamic trend energy value.

[0027] As a preferred embodiment of the present invention, step S5 involves calculating the basic probability score configuration confidence and performing multi-node orthogonal fusion based on DS evidence theory, including:

[0028] Substituting the absolute dynamic trend energy value and the corresponding decoupling scalar eigenvalue into the dimensionless normalized exponential penalty model, the basic probability assignment values ​​of the electrical sensing node and the acoustic sensing node for the abnormal target event A are calculated. ,satisfy:

[0029] ;

[0030] in, These correspond to electrical and acoustic fields, respectively. and These correspond to the static deviation penalty weight coefficient and the dynamic energy penalty weight coefficient, respectively. This represents the baseline value for a healthy state; Represents the dynamic energy normalization constant; This represents the decoupled scalar eigenvalue at the current moment; Indicates the energy value representing the dynamic trend over time; This represents an exponential function with the natural constant e as its base.

[0031] normal events The corresponding basic probability assignment values ​​satisfy By invoking the orthogonal operators of the DS evidence theory to cross-fuse the confidence scores of the two nodes, the final decision confidence probability of fluid anomaly at the current moment is solved. ,satisfy:

[0032] ;

[0033] in, and These represent the probability assignment values ​​of the electrical and acoustic feature components to the abnormal target event, respectively. and These represent the probability assignment values ​​of the electrical and acoustic characteristic components to normal events, respectively.

[0034] As a preferred embodiment of the present invention, step S6, generating the comprehensive degradation index of the electrolyte fluid, includes:

[0035] The absolute dynamic trend energy values ​​of the electrical and acoustic channels are dimensionless using the corresponding dynamic energy normalization constants, and then weighted and summed using the corresponding degradation sensitivity weighting coefficients.

[0036] The weighted summation result is added to the natural constant, and then a logarithmic operation is performed. The logarithmic result is multiplied by the final decision confidence probability to obtain the electrolyte fluid comprehensive deterioration index.

[0037] As a preferred embodiment of the present invention, before step S1, the method further includes:

[0038] The peripheral memory interface FMC sends an addressing instruction to the built-in ROM, reads the global parameter matrix and system configuration parameters obtained from the pre-offline calibration, and writes them into the global variable resident area of ​​the static random access memory SRAM. The built-in hardware CRC32 module is then called to perform cyclic redundancy check. After the check passes, a ready level is sent to the underlying communication bus to enable the multi-channel physical sampling hardware interrupt sequence.

[0039] Secondly, this disclosure proposes an intelligent monitoring system for electrolyte fluid based on multimodal sensing, wherein the system specifically includes the following logical functions:

[0040] The parameter initialization and loading module is used to read the global parameter matrix and system configuration parameters obtained from offline calibration after power-on and write them into the static random access memory (SRAM). After the hardware cyclic redundancy check passes, the hardware interrupt sequence is enabled.

[0041] The multimodal physical signal acquisition module is used to drive the multimodal physical sensing matrix to perform low-level signal synchronous acquisition, encapsulate the multi-source physical column vector sequence, and push it into the L1 cache through direct memory access controller DMA zero copy.

[0042] The thermodynamic baseline compensation module is used to perform thermodynamic baseline compensation on the multi-source physical column vector sequence, generate pure multimodal features and write them into the compensation feature area of ​​SRAM, and flip the baseline compensation completion flag bit of the status register.

[0043] The orthogonal decoupling and dimensionality reduction pooling module is used to concatenate and stitch pure multimodal features into a multimodal feature column vector, perform spatial orthogonal feature decoupling and high-dimensional feature scalar pooling compression mapping, output pure integrated electrical feature components and independent acoustic scalar feature components, and push them into the first-in-first-out (FIFO) buffer queue;

[0044] The temporal high-order differential mapping module is used to extract temporal feature data from the FIFO buffer queue based on pointer addressing, perform temporal high-order differential mapping, extract the absolute dynamic trend energy value and write it into the multi-node evidence register group;

[0045] The multi-node evidence fusion decision module is used to read the data in the multi-node evidence register group, substitute it into the dimensionless normalized exponential penalty model to calculate the basic probability score configuration confidence, call the DS evidence theory orthogonal and operator to eliminate node conflicts, solve the final decision confidence probability and latch it into the TrustZone security isolation area in real time.

[0046] The integrated degradation assessment and adaptive control module is used to read the decision probability and dynamic energy in the register group within the isolation zone, perform engineering fusion mapping to generate the electrolyte fluid integrated degradation index, and perform hardware branch judgment on the electrolyte fluid integrated degradation index, the first warning threshold and the second protection threshold to drive the external execution terminal to execute the corresponding hierarchical interlocking protection action.

[0047] As a preferred embodiment of the present invention, the system includes a main control chip, which integrates the following functional units:

[0048] The central processing unit (CPU) is used to coordinate data parsing, vector encapsulation, and system state machine scheduling.

[0049] The floating-point unit (FPU) is used to respond to calls from the thermodynamic baseline compensation module and perform impedance spectrum temperature compensation calculations and ultrasonic attenuation temperature and pressure joint compensation calculations.

[0050] The Tensor Accelerator (NPU) is used to respond to calls to the orthogonal decoupling and dimensionality reduction pooling modules, and to perform autocorrelation similarity matrix calculation, eigenvector decoupling, and Euclidean norm pooling compression mapping.

[0051] The digital signal processing (DSP) core has a built-in digital differential filter, which is used to respond to the call of the time-series high-order differential mapping module and perform time-series feature data extraction and high-order differential mapping.

[0052] The logic reasoning engine is used to respond to calls from the multi-node evidence fusion decision module and execute confidence calculation and cross-fusion operators.

[0053] As a preferred embodiment of the present invention, the system further includes a multimodal physical sensing matrix, a memory, communication peripherals, and an external execution terminal, wherein:

[0054] The multimodal physical sensing matrix includes a reference electrode-free miniature impedance measurement probe, an ultrasonic transceiver array, and a miniature integrated temperature and pressure sensor.

[0055] The memory includes read-only memory (ROM) and static random access memory (SRAM);

[0056] The main control chip also integrates a peripheral memory interface (FMC), a hardware CRC32 verification module, a direct memory access controller (DMA), a direct digital frequency synthesizer (DDS), an acoustic drive circuit, an advanced peripheral bus (APB), a status register, a receive register, a transmit holding register, a bus mapping unit, a multi-node evidence register group, and a TrustZone security isolation zone.

[0057] External execution terminals include programmable logic controllers (PLCs), on-site audible and visual alarms, fluid circulation pumps, pipeline solenoid valves, and main circuit contactors.

[0058] In summary, the present invention has the following beneficial effects:

[0059] Firstly, to address the failure of optical vision technology caused by liquid turbidity and bubble refraction, this invention employs a multimodal physical sensing matrix composed of a reference electrode-less micro impedance measurement probe and an ultrasonic transceiver array to acquire the underlying signal. Simultaneously, to address the interference caused by fluid temperature changes and pressure pulsations, this invention calls upon the floating-point unit (FPU) to perform impedance spectrum temperature compensation and ultrasonic attenuation temperature and pressure joint compensation, achieving thermodynamic baseline flattening at the hardware level. This ensures the purity of the characteristic signal from the physical source and significantly reduces system false alarms caused by thermodynamic drift.

[0060] Secondly, to address the overlapping observation problem between electrochemical impedance and ultrasonic attenuation in characterizing phenomena such as bubble precipitation, this invention calculates the autocorrelation similarity matrix using a tensor accelerator (NPU) and subtracts the identity matrix. After normalization, an orthogonal feature decoupling weight matrix is ​​generated, and a forced inverse mapping is performed to remove the physical overlap components across modes. At the same time, by performing Euclidean norm pooling compression mapping on the high-dimensional electrical feature vectors, the tensor morphological dimension of the multimodal data is unified. This avoids the secondary amplification of anomalous features in multiple sensors and significantly reduces the memory usage of subsequent time-series operations.

[0061] Thirdly, addressing the issue that static characteristics at a single moment are easily masked by the slow, natural decay of the electrolyte over a long period, this invention employs a digital differential filter to perform temporal high-order differential mapping on the pure scalar characteristics, accurately extracting the absolute dynamic trend energy value. This mechanism enables the system to keenly detect the evolution rate and acceleration of fluid anomalies before a fault causes macroscopic exceedances, gaining crucial time lead time to prevent severe deterioration.

[0062] Fourth, addressing the issue of electromagnetic interference in industrial settings easily causing false shutdowns of single nodes, this invention substitutes independent electrical and acoustic sensing nodes into a dimensionless normalized exponential penalty model, quantifies the basic probability distribution reliability, and uses orthogonal operators from the DS evidence theory for cross-fusion. By normalizing the conflict coefficient of the denominator, the outlier characteristics of single nodes are effectively suppressed, and a highly reliable comprehensive anomaly decision probability is output only when two nodes coordinate an alarm.

[0063] Fifth, addressing the problem that complex probability matrices cannot directly drive external industrial equipment, this invention performs logarithmic coupling engineering fusion mapping, transforming decision probabilities and dynamic energy into a dimensionless comprehensive degradation index with excellent linearity across the entire range. By comparing the state machine hardware branches with the preset first warning threshold and second protection threshold in the static random access memory (SRAM), the system can adaptively issue instructions to external terminals, achieving graded interlocking protection from power reduction audible and visual alarms to hard interrupt locking of solenoid valves and main circuit contactors. Attached Figure Description

[0064] Figure 1 A flowchart illustrating an intelligent monitoring method for electrolyte fluid based on multimodal sensing, provided as an embodiment of the present invention;

[0065] Figure 2 This is a framework diagram of an intelligent monitoring system for electrolyte fluid based on multimodal sensing, provided in an embodiment of the present invention. Detailed Implementation

[0066] The present application will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present application, but do not limit the present application in any way. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of the present application. These all fall within the protection scope of the present application.

[0067] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0068] Unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The term "and / or" as used in this specification includes any and all combinations of one or more of the associated listed items.

[0069] Furthermore, the technical features involved in the various embodiments of this application described below can be combined with each other as long as they do not conflict with each other.

[0070] This disclosure aims to address the problems of traditional electrolyte fluid monitoring methods' over-reliance on optical or visual imaging technologies, which leads to a significant decrease in monitoring accuracy under harsh industrial conditions such as liquid turbidity, bubble refraction, and light attenuation. Furthermore, single-modal sensors are highly susceptible to thermodynamic drift and hardware noise interference, resulting in high false alarm rates. Therefore, this disclosure proposes a multimodal sensing-based intelligent electrolyte fluid monitoring method and system for robustly monitoring and assessing the deterioration and health status of electrolyte fluids in complex and harsh industrial environments. This method and system employ a purely physical multi-source sensing matrix and chain-like mathematical computation architecture that completely eliminates optical vision. Through underlying hardware and software collaborative mechanisms such as thermodynamic baseline compensation, spatial orthogonal feature decoupling, temporal high-order differential mapping, and DS multi-node evidence theory fusion, it effectively eliminates complex and variable physical interference and cross-modal redundancy in industrial settings. This achieves the goals of accurately capturing early microscopic anomalies, completely suppressing single-node false alarms, and significantly improving the reliability of system engineering implementation.

[0071] The hardware architecture relied upon in this embodiment includes: a multimodal physical sensing matrix, a main control chip, a memory, communication peripherals, and an external execution terminal.

[0072] The multimodal physical sensing matrix includes: a reference electrode-less miniature impedance measurement probe with electrodes made of corrosion-resistant inert metals such as platinum and an electrode spacing fixed between 0.5 mm and 2.0 mm; an ultrasonic transceiver array with a center frequency configured in the 1 MHz to 10 MHz band; and a miniature integrated temperature and pressure sensor with a measurement accuracy of 0.1 grade and a pressure range of 0-2 MPa.

[0073] The main control chip integrates the following components: a central processing unit (CPU), a floating-point unit (FPU), a tensor accelerator (NPU), a direct memory access controller (DMA), a direct digital frequency synthesizer (DDS), an acoustic drive circuit, a digital signal processing (DSP) kernel, a logic inference engine, a comprehensive degradation assessment module, a hardware CRC32 check module, a TrustZone security isolation zone, a multi-node evidence register group, an advanced peripheral bus (APB), a peripheral memory interface (FMC), a status register, a receive register, a transmit holding register, and a bus mapping unit.

[0074] Memory includes read-only memory (ROM) and static random access memory (SRAM).

[0075] External execution terminals include programmable logic controllers (PLCs), on-site audible and visual alarms, fluid circulation pumps, pipeline solenoid valves, and main circuit contactors.

[0076] Please refer to Figure 1 , Figure 1A flowchart of an intelligent monitoring method for electrolyte fluid based on multimodal sensing, as described in an embodiment of this disclosure, is shown, including the following steps:

[0077] Step S0: System power-on initialization and centralized loading of offline parameters.

[0078] S0.1, Perform hardware addressing and read global parameters and symbol definitions;

[0079] After the system is powered on, the CPU of the main control chip sends an addressing instruction to the built-in ROM through the FMC interface to read the global parameter matrix and system configuration parameters that were obtained in advance under a standard laboratory environment of 25°C and 1 standard atmosphere by repeatedly calibrating and fitting with the least squares method using a standard electrochemical workstation and an ultrasonic calibration system.

[0080] The global parameter matrix includes: sampling period Number of discrete excitation frequency points and frequency vector Basic activation energy of electrolyte Ideal gas constant R; Standard reference absolute temperature Standard reference absolute pressure Temperature first-order sensitivity compensation coefficient First-order pressure sensitivity compensation coefficient First-order time-difference gain coefficient Second-order time-difference gain coefficient ; Healthy state normal baseline value and Dynamic energy normalization constant and Static deviation penalty weight coefficient and Dynamic energy penalty weighting coefficient and Degradation sensitivity weighting coefficient and First warning threshold With the second protection threshold .

[0081] The system configuration parameters include: the DDS excitation waveform type is configured as a sine wave; the analog-to-digital converter (ADC) sampling rate is configured as one megahertz per second; the CRC32 check adopts the CRC32-IEEE802.3 standard polynomial; and the first-in-first-out buffer queue (FIFO) depth is set to eight levels.

[0082] In this embodiment, the general mathematical operators for system-level calls are uniformly defined as follows: The transpose operation represents a matrix or vector; This indicates the computational norm of a vector; Represents the identity matrix of the corresponding dimension; This represents an exponential function with the natural constant e as its base. This represents a logarithmic function with the natural constant e as its base. This indicates that normalization mapping is performed according to the matrix row dimensions; This represents absolute value operations.

[0083] S0.2, Enable Execution Parameter Resident and Sampling Interrupt;

[0084] The CPU reads the global parameter matrix and system configuration parameters and writes them into the global variable resident area of ​​SRAM. It then calls the built-in hardware CRC32 module to perform cyclic redundancy check. After the check passes, the CPU sends a ready level to the underlying communication bus to enable the multi-channel physical sampling hardware interrupt sequence.

[0085] Step S1: Drive the multimodal physical sensing matrix to perform low-level signal synchronous acquisition, and obtain and encapsulate a multi-source physical column vector sequence containing synchronization timestamps.

[0086] S1.1, Drive the multimodal physical sensing matrix to perform low-level signal acquisition;

[0087] The DDS inside the main control chip generates a swept-frequency sinusoidal excitation waveform based on the frequency vector in SRAM, drives the micro impedance measurement probe to inject signals and acquire the electrochemical impedance response; simultaneously, the acoustic drive circuit excites the ultrasonic transceiver array to emit and acquire ultrasonic echoes; the temperature and pressure integrated sensor acquires thermodynamic parameters synchronously through the ADC channel.

[0088] S1.2 Parses the data packet and performs initialization vector encapsulation;

[0089] The main control chip's receive register captures data packets from each sensor. After the CPU removes the frame header and checksum, it encapsulates the payload into a multi-source physical column vector sequence containing a synchronization timestamp t. Specifically, this sequence includes the original dynamic electrochemical impedance spectral amplitude vector. Original ultrasonic attenuation coefficient Real-time absolute temperature and real-time absolute pressure .

[0090] S1.3 triggers DMA zero-copy and data handover;

[0091] The encapsulated column vector sequence is pushed into the main control chip's L1 cache via a zero-copy DMA channel. The DMA controller then sends a hardware interrupt pulse to the CPU, waking up the FPU to enter the arithmetic logic operation flow corresponding to step S2.

[0092] Step S2: Call the dynamic compensation algorithm module to perform thermodynamic baseline compensation on the multi-source physical column vector sequence, generate pure multimodal features and write them into the static random access memory (SRAM).

[0093] S2.1 Perform impedance spectrum temperature compensation calculation;

[0094] The electrolyte ion migration rate increases exponentially with increasing temperature, leading to baseline impedance drift. The FPU retrieves the original impedance spectrum amplitude vector and real-time absolute temperature from the L1 cache. Perform hardware-level baseline compensation; for each excitation frequency point The following compensation formula is applied:

[0095] ;

[0096] in, This represents the dynamic impedance amplitude after temperature compensation. This represents the original dynamic electrochemical impedance spectral amplitude of the S1.2 package; Indicates the basic activation energy of the electrolyte; Represents the ideal gas constant; This indicates the real-time absolute temperature of the fluid. Indicates the standard reference absolute temperature; This represents the excitation frequency point, where k is a positive integer and... .

[0097] S2.2 Perform ultrasonic attenuation temperature and pressure joint compensation calculation;

[0098] The scattering and attenuation of ultrasound waves in fluids are directly affected by fluid density, which is determined by both temperature and pressure; the FPU retrieves the original ultrasound attenuation coefficient. and real-time absolute pressure Perform combined temperature and pressure baseline compensation using the following formula:

[0099] ;

[0100] in, This represents the pure ultrasonic attenuation coefficient after temperature and pressure compensation. This represents the original ultrasonic wave attenuation coefficient; and These represent the first-order sensitivity compensation coefficients for temperature and pressure, respectively. Indicates the real-time absolute pressure of the fluid; This indicates the standard reference absolute pressure.

[0101] S2.3, Perform memory cleanup and accelerator wake-up;

[0102] Limited by the L1 cache resources of the embedded system, after the FPU completes the calculation, the CPU schedules the DMA controller to release the cache address space occupied by the real-time absolute temperature and real-time absolute pressure; the FPU writes the pure ultrasonic attenuation coefficient and the dynamic impedance amplitude at each frequency point into the compensation feature area of ​​the SRAM, and sets the baseline compensation completion flag of the status register to high, and selects the tensor accelerator NPU through the APB bus to enter step S3.

[0103] Step S3: Construct feature column vectors in the tensor space and perform orthogonal decoupling and dimensionality reduction pooling to output independent integrated electrical feature components and independent acoustic scalar feature components, and push them into the first-in-first-out (FIFO) buffer queue.

[0104] S3.1. Allocate contiguous memory addresses and concatenate multimodal feature column vectors;

[0105] Electrochemical impedance and ultrasonic attenuation exhibit physical collinearity overlap when characterizing phenomena such as bubble precipitation. To provide a computational platform for redundancy stripping, the NPU concatenates and splices the pure features of each frequency band into a multimodal feature column vector within a contiguous memory address block, executing the following formula:

[0106] ;

[0107] in, Indicates the current time The concatenated multimodal feature column vector has a dimension of [dimensionality missing]. .

[0108] S3.2 Calculate the spatial orthogonal feature decoupling weight matrix;

[0109] The NPU calls a matrix multiplier to calculate the autocorrelation similarity matrix of the feature column vectors. The diagonal elements of the normalized autocorrelation matrix are always 1, corresponding to the absolute similarity between each feature component and itself. To highlight the cross-modal cross-correlation, the NPU subtracts the identity matrix. Set the diagonal to zero, remove its own similarity components, and use the remaining cross-correlation terms as the penalty basis to perform row-by-row processing. Normalize and generate decoupling weights by executing the following formula:

[0110] ;

[0111] in, The cross-modal redundancy penalty orthogonal feature decoupling weight matrix has dimensions of . ; express The transpose of ; express The second norm; The dimension is The identity matrix.

[0112] S3.3, Mapping to generate decoupled fluid feature column vectors;

[0113] The NPU applies the decoupling weight matrix inversely to the original feature column vector, forcibly stripping away physically overlapping components to obtain mutually independent modal features, as shown in the following formula:

[0114] ;

[0115] in, This represents the clean feature column vector after eliminating cross-modal physical redundancy, with dimension . The NPU then intercepted... The former Each element serves as a high-dimensional electrical feature vector. Extract the last element as an independent acoustic scalar feature component. .

[0116] S3.4 Perform high-dimensional feature scalarization pooling compression;

[0117] To unify the tensor shape dimension of multimodal data, align high-dimensional electrical vectors with one-dimensional acoustic scalars, and reduce the memory overhead of subsequent temporal operations, the NPU performs Euclidean norm pooling compression mapping on the high-dimensional electrical feature vectors, using the following formula:

[0118] ;

[0119] in, represents the pure comprehensive electrical characteristic components after pooling and compression; n corresponds to the number of discrete excitation frequency points. This represents the k-th component element of a high-dimensional electrical eigenvector.

[0120] S3.5, Data queuing and hardware switching;

[0121] NPU will integrate pure electrical characteristic components of the same dimension. With independent acoustic scalar characteristic components Push the head of the FIFO buffer queue dedicated to the DSP core into the queue; the offset update action of the head pointer triggers a falling edge hardware interrupt, activating the digital differential filter in the DSP and proceeding to step S4.

[0122] Step S4: Call the digital differential filter to extract the feature component time series data in the FIFO buffer queue according to the pointer address, perform high-order time series mapping, and extract the absolute dynamic trend energy value.

[0123] S4.1 Extract timing feature data based on pointer addressing;

[0124] Early core characteristics of electrolyte degradation are manifested in abnormal evolution rate and evolution acceleration; the digital differential filter reads the scalar features of the current moment through the head pointer and traverses the tail pointer according to the address offset to retrieve the historical moment features of the corresponding period from the FIFO deep cache queue in order to filter out long-term slow natural decay interference.

[0125] S4.2 Execute the higher-order difference mapping equation;

[0126] The digital differential filter maps the pure synthesized electrical characteristic components and the independent acoustic scalar characteristic components using differential operators to extract the absolute dynamic trend energy value. :

[0127]

[0128] in, Indicates the first Temporal dynamic trend energy value of each physical characteristic component; The set of values ​​contains and These correspond to electricity and acoustics, respectively; and These correspond to the first-order time-difference gain coefficient and the second-order time-difference gain coefficient, respectively. Corresponding sampling period; This represents the decoupled scalar eigenvalue at the current moment when it is pushed into the head of the FIFO queue; and These represent the corresponding historical feature values ​​read from the FIFO by the address offset from the previous sampling period and twice the sampling period ago, respectively.

[0129] S4.3, Packaging and Authorization of Multi-Node Evidence Registers;

[0130] The DSP core will output the time-series dynamic trend energy value Decoupled scalar eigenvalues ​​from the current time step The data is written to the multi-node evidence register group configured in the hardware via the data bus; the register group sends an acknowledgment handshake signal to the logic inference engine in the CPU, authorizing it to read-only access and proceeding to step S5.

[0131] Step S5: Substitute the absolute dynamic trend energy value and the corresponding decoupled scalar feature value into the dimensionless normalized exponential penalty model, calculate the basic probability distribution confidence, perform multi-node orthogonal fusion based on DS evidence theory, solve the comprehensive abnormal decision probability and lock it to the TrustZone security isolation zone in real time.

[0132] S5.1 Calculate the basic probability score configuration confidence by calling the exponential penalty model;

[0133] To suppress false alarms caused by environmental noise in a single sensor channel, the logic inference engine sets the abnormal degradation of the electrolyte as target event A. The inference engine reads feature scalars and dynamic energy from the multi-node evidence register set, substitutes them into a dimensionless normalized exponential penalty model, and quantifies the basic probability allocation values ​​of the two types of independent sensor nodes for the abnormal target event:

[0134] ;

[0135] in, Indicates the first Each feature component supports the basic probability assignment value of target event A; The set of values ​​contains and These correspond to electricity and acoustics, respectively; and These correspond to the static deviation penalty weight coefficient and the dynamic energy penalty weight coefficient, respectively. This represents the baseline value for a healthy, normal state, used to perform dimensionless transformation on static characteristics. This represents the dynamic energy normalization constant, used to perform dimensionless transformation on dynamic characteristics.

[0136] The logic reasoning engine synchronously calculates the corresponding normal events. The basic probability distribution values ​​satisfy the equation .

[0137] S5.2 Execute the orthogonal fusion operator to eliminate node conflicts;

[0138] The logical reasoning engine calls the orthogonal operators of the DS evidence theory to cross-fuse the confidence scores of two nodes, and normalizes the conflict coefficient of the denominator to suppress outlier features of a single node, executing the following formula:

[0139] ;

[0140] in, This represents the credible probability of the final decision regarding the occurrence of fluid anomalies at the current moment, after joint evidence conflict suppression. and These represent the probability assignment values ​​of the electrical and acoustic feature components to the abnormal target event, respectively. and These represent the probability assignment values ​​of the electrical and acoustic characteristic components to normal events, respectively.

[0141] S5.3, Safe Zone Latching and Soft Interrupt Triggering;

[0142] The final decision reliability probability is latched in real time to the TrustZone security isolation area by the CPU bus instruction to prevent illegal overwriting; the latch instruction triggers the bus mapping unit to generate an advanced soft interrupt, which transfers control to the comprehensive degradation evaluation module inside the CPU and proceeds to step S6.

[0143] Step S6: Execute engineering fusion mapping to generate a full-range comprehensive degradation index, and perform hardware branch determination on the comprehensive degradation index, the first warning threshold and the second protection threshold, and drive the external terminal to execute adaptive interlocking protection actions.

[0144] S6.1 Execute the engineering fusion mapping to generate the degradation index;

[0145] To provide external PLCs with dimensionless control protocol messages exhibiting excellent linearity across the entire range, the comprehensive degradation assessment module reads the decision probabilities from TrustZone and the dynamic energy from the register set, performs engineering fusion mapping, and generates a comprehensive degradation index.

[0146] ;

[0147] in, This indicates the overall degradation index of the electrolyte fluid being output. This represents the final decision probability calculated in S5.2; This represents a logarithmic function; e is a natural constant, ensuring that the argument is not less than e to avoid negative bias results; This represents the degradation sensitivity weighting coefficient retrieved from SRAM.

[0148] S6.2 State machine comparison and hardware actuator drive instruction issuance;

[0149] The comprehensive degradation assessment module compares the fluid comprehensive degradation index with the first warning threshold in the SRAM. Second protection threshold Perform hardware branch determination:

[0150] (1) If The state machine maintains a safe steady-state flag and executes step S6.3;

[0151] (2) If The main control chip pushes an alarm message into the transmit holding register (TXHoldingRegister), which is then sent to the PLC via the communication peripherals to drive the on-site audible and visual alarm and lower the control level of the fluid circulation pump.

[0152] (3) If The main control chip triggers the highest level unmasked interrupt (NMI) in the hardware, directly outputting a high-level pulse to the general purpose input / output (GPIO) pin, controlling the main circuit contactor to open and lock the pipeline solenoid valve.

[0153] S6.3, Bus pointer release and system cycle sleep;

[0154] After instruction comparison is completed, the CPU of the main control chip sends a reset pulse to clear the temporary status flags of the FIFO queue and DMA cache for the current cycle and reset the read / write pointers; the system clock frequency is reduced to enter a low-power sleep mode, waiting for the next sampling cycle. The clock wakes you up at set intervals to execute the next round of automatic monitoring closed loop.

[0155] This disclosure also proposes an intelligent monitoring system for electrolyte fluid based on multimodal sensing. The system's logical functional architecture corresponds to the intelligent monitoring method for electrolyte fluid based on multimodal sensing provided in the foregoing embodiments of this application. Each functional module is a logical functional unit implemented based on underlying hardware entities. The underlying hardware entities upon which this system relies include: a multimodal physical sensing matrix, a main control chip, a memory, communication peripherals, and an external execution terminal. The multimodal physical sensing matrix includes a reference electrode-less miniature impedance measurement probe, an ultrasonic transceiver array, and a miniature integrated temperature and pressure sensor. The main control chip includes a central processing unit (CPU), a floating-point unit (FPU), a tensor accelerator (NPU), a direct memory access controller (DMA), a direct digital frequency synthesizer (DDS), a digital signal processing (DSP) core, and a logic inference engine, among other hardware acceleration units. The external execution terminal includes a programmable logic controller (PLC), a field audible and visual alarm, a fluid circulation pump, pipeline solenoid valves, and a main circuit contactor.

[0156] Please refer to Figure 2 , Figure 2 A framework diagram of an intelligent electrolyte fluid monitoring system based on multimodal sensing, as described in an embodiment of this disclosure, is shown. The system specifically includes, in terms of logical functions, a parameter initialization and loading module, a multimodal physical signal acquisition module, a thermodynamic baseline compensation module, an orthogonal decoupling and dimensionality reduction pooling module, a temporal high-order differential mapping module, a multi-node evidence fusion decision module, and a comprehensive degradation assessment and adaptive control module.

[0157] The parameter initialization and loading module is used to send addressing instructions to the built-in read-only memory ROM via the peripheral memory interface FMC after the system is powered on, read the global parameter matrix and system configuration parameters obtained through pre-offline calibration, and write the above parameters into the global variable resident area of ​​the static random access memory SRAM. After calling the hardware CRC32 module to complete the cyclic redundancy check, it enables the multi-channel physical sampling hardware interrupt sequence. The specific parameter definition, data allocation and interrupt triggering logic of this module are completely consistent with the description of step S0 in the aforementioned method embodiment.

[0158] The multimodal physical signal acquisition module drives the multimodal physical sensing matrix to perform low-level signal synchronous acquisition. It generates a swept-frequency sinusoidal excitation waveform through a direct digital frequency synthesizer (DDS) to synchronously drive impedance measurement, ultrasonic acquisition, and thermodynamic parameter sampling. The parsed data is encapsulated into a multi-source physical column vector sequence containing synchronization timestamps and pushed into the main control chip's L1 cache via a direct memory access controller (DMA) with zero-copy propagation. The specific waveform excitation rules, vector encapsulation rules, and data transfer mechanism of this module are completely consistent with the description in step S1 of the aforementioned method embodiment.

[0159] The thermodynamic baseline compensation module calls the floating-point unit (FPU) to perform impedance spectrum temperature compensation and ultrasonic attenuation temperature-pressure joint compensation calculations based on the extracted real-time absolute temperature and pressure. This generates pure multimodal features and writes them into the compensation feature area of ​​the SRAM. Upon completion, the baseline compensation completion flag in the status register is flipped. The specific compensation formula, thermodynamic parameter calling logic, and flag flipping mechanism of this module are completely consistent with step S2 described in the aforementioned method embodiment.

[0160] The orthogonal decoupling and dimensionality reduction pooling module calls the tensor accelerator NPU to concatenate and stitch the pure multimodal features into a multimodal feature column vector, calculates the autocorrelation similarity matrix, and removes its own similarity components to generate an orthogonal feature decoupling weight matrix. This weight matrix is ​​then applied inversely to the original feature column vector to remove physically overlapping components. Euclidean norm pooling compression mapping is performed on the truncated high-dimensional electrical feature vector. Finally, the pure integrated electrical feature components of the same dimension and the independent acoustic scalar feature components are pushed into a dedicated first-in-first-out (FIFO) buffer queue. The specific decoupling operation formula, pooling compression formula, and data enqueueing mechanism of this module are completely consistent with the description in step S3 of the aforementioned method embodiment.

[0161] The temporal higher-order differential mapping module is used to invoke the digital differential filter, extract the temporal feature data from the FIFO buffer queue according to pointer addressing, and call the differential operators to perform temporal higher-order differential mapping to extract the absolute dynamic trend energy value. This value, along with the corresponding decoupled scalar feature value, is then written into the multi-node evidence register group. The specific higher-order differential mapping equation, natural decay interference filtering logic, and register authorization mechanism of this module are completely consistent with the description in step S4 of the aforementioned method embodiment.

[0162] The multi-node evidence fusion decision module is used to call the logic inference engine to read data from the multi-node evidence register group, substitute it into the dimensionless normalized exponential penalty model to calculate the basic probability distribution confidence of independent sensor nodes, and call the DS evidence theory orthogonal operator to eliminate node conflicts, solve for the comprehensive anomaly decision probability, and latch it into the TrustZone security isolation area in real time. The specific penalty model formula, orthogonal fusion operator logic, and soft interrupt triggering mechanism of this module are completely consistent with the description of step S5 in the aforementioned method embodiment.

[0163] The integrated degradation assessment and adaptive control module reads the decision probability within the isolation zone and the dynamic energy in the register group, performs engineering fusion mapping to generate a full-range electrolyte fluid integrated degradation index, and compares this index with the first warning threshold and the second protection threshold in SRAM through hardware branching. It then adaptively drives the external execution terminal to execute the corresponding hierarchical interlocking protection actions. The specific fusion mapping formula, state machine comparison logic, and hardware actuator drive instruction issuance mechanism of this module are completely consistent with the description in step S6 of the aforementioned method embodiment.

[0164] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for intelligent monitoring of electrolyte fluid based on multimodal sensing, applied to a main control chip, characterized in that, include: S1. Low-level signal synchronous acquisition and encapsulation: Drive the multi-modal physical sensing matrix to perform low-level signal synchronous acquisition, obtain and encapsulate a multi-source physical column vector sequence containing synchronization timestamps; S2. Thermodynamic baseline compensation: The dynamic compensation algorithm module is called to perform thermodynamic baseline compensation on the multi-source physical column vector sequence, generate pure multimodal features and write them into the compensation feature area of ​​static random access memory (SRAM). S3, Orthogonal Decoupling and Dimensional Reduction Pooling: Construct feature column vectors in the tensor space and perform orthogonal decoupling and dimensional reduction pooling to output pure integrated electrical feature components and independent acoustic scalar feature components, and push them into the first-in-first-out (FIFO) buffer queue; S4, Temporal Higher-Order Differential Mapping: Call the digital differential filter to extract the feature component temporal data in the FIFO buffer queue according to pointer addressing, perform temporal higher-order differential mapping, and extract the absolute dynamic trend energy value; S5. Multi-node evidence fusion decision: Substitute the absolute dynamic trend energy value and the corresponding decoupled scalar feature value into the dimensionless normalized exponential penalty model, calculate the basic probability distribution confidence, perform multi-node orthogonal fusion according to DS evidence theory, solve the final decision credibility probability and lock it to the TrustZone security isolation area in real time. S6. Comprehensive Deterioration Assessment and Adaptive Control: Execute engineering fusion mapping to generate a comprehensive deterioration index of electrolyte fluid, and perform hardware branch judgment on the comprehensive deterioration index of electrolyte fluid with the first warning threshold and the second protection threshold to drive the external terminal to execute adaptive interlocking protection actions; Step S3 performs orthogonal decoupling and dimensionality reduction pooling, including: The pure multimodal features are concatenated into a multimodal feature column vector in contiguous memory address blocks. Calculate its autocorrelation similarity matrix by subtracting the identity matrix. Remove its own similarity components and perform row-by-row processing. Normalization generates cross-modal redundancy penalty orthogonal feature decoupling weight matrix ,satisfy: ； in, express The transpose of ; express The second norm; This indicates that normalization mapping is performed according to the matrix row dimensions; Represents the identity matrix of the corresponding dimension; Indicates the current sampling time; Will Reverse action on the After stripping away the physically overlapping components, a decoupled fluid feature column vector is generated. , cut off the The former Each element serves as a high-dimensional electrical feature vector. and to Perform Euclidean norm pooling compression mapping to output pure synthesized electrical eigencomponents. ,satisfy: ； in, The corresponding number of discrete excitation frequency points; This represents the k-th component element of a high-dimensional electrical eigenvector; The last element is used as an independent acoustic scalar characteristic component. .

2. The intelligent monitoring method for electrolyte fluid based on multimodal sensing according to claim 1, characterized in that, Step S2 involves thermodynamic baseline compensation of the multi-source physical column vector sequence, including: Based on the real-time absolute temperature of the fluid, the preset standard reference absolute temperature, and the basic activation energy of the electrolyte, the original dynamic electrochemical impedance spectrum amplitude is subjected to exponential temperature compensation. Based on the first-order change of the fluid's real-time absolute temperature relative to the standard reference absolute temperature and the first-order change of the fluid's real-time absolute pressure relative to the standard reference absolute pressure, and combined with the corresponding sensitivity compensation coefficient, the original ultrasonic attenuation coefficient is linearly compensated for temperature and pressure.

3. The intelligent monitoring method for electrolyte fluid based on multimodal sensing according to claim 1, characterized in that, Step S4 extracts the absolute dynamic trend energy value, including: The current scalar feature is read by the head pointer, and the historical feature of the corresponding period is retrieved by traversing the tail pointer according to the address offset. The time series higher-order difference mapping is executed by calling the difference operator respectively. The first-order trend energy is determined based on the product of the first-order time difference gain coefficient and the absolute value of the first-order characteristic rate of change. The second-order trend energy is determined based on the product of the second-order time difference gain coefficient and the absolute value of the second-order characteristic rate of change. The first-order trend energy and the second-order trend energy are weighted and summed to obtain the corresponding absolute dynamic trend energy value.

4. The intelligent monitoring method for electrolyte fluid based on multimodal sensing according to claim 1, characterized in that, Step S5 calculates the basic probability score configuration confidence and performs multi-node orthogonal fusion based on DS evidence theory, including: Substituting the absolute dynamic trend energy value and the corresponding decoupling scalar eigenvalue into the dimensionless normalized exponential penalty model, the basic probability assignment values ​​of the electrical sensing node and the acoustic sensing node for the abnormal target event A are calculated. ,satisfy: ； in, These correspond to electrical and acoustic fields, respectively. and These correspond to the static deviation penalty weight coefficient and the dynamic energy penalty weight coefficient, respectively. This represents the baseline value for a healthy state; Represents the dynamic energy normalization constant; This represents the decoupled scalar eigenvalue at the current moment; Indicates the energy value representing the dynamic trend over time; This represents an exponential function with the natural constant e as its base. normal events The corresponding basic probability assignment values ​​satisfy By invoking the orthogonal operators of the DS evidence theory to cross-fuse the confidence scores of the two nodes, the final decision confidence probability of fluid anomaly at the current moment is solved. ,satisfy: ； in, and These represent the probability assignment values ​​of the electrical and acoustic feature components to the abnormal target event, respectively. and These represent the probability assignment values ​​of the electrical and acoustic characteristic components to normal events, respectively.

5. The intelligent monitoring method for electrolyte fluid based on multimodal sensing according to claim 1, characterized in that, Step S6 generates the overall degradation index of the electrolyte fluid, including: The absolute dynamic trend energy values ​​of the electrical and acoustic channels are respectively processed into dimensionless values ​​using the corresponding dynamic energy normalization constants, and then weighted and summed using the corresponding degradation sensitivity weighting coefficients. The weighted summation result is added to the natural constant, and then a logarithmic operation is performed. The logarithmic result is multiplied by the final decision confidence probability to obtain the electrolyte fluid comprehensive deterioration index.

6. The intelligent monitoring method for electrolyte fluid based on multimodal sensing according to claim 1, characterized in that, Before step S1, the following is also included: The peripheral memory interface FMC sends an addressing instruction to the built-in ROM, reads the global parameter matrix and system configuration parameters obtained from the pre-offline calibration, and writes them into the global variable resident area of ​​the static random access memory SRAM. The built-in hardware CRC32 module is then called to perform cyclic redundancy check. After the check passes, a ready level is sent to the underlying communication bus to enable the multi-channel physical sampling hardware interrupt sequence.

7. An intelligent monitoring system for electrolyte fluid based on multimodal sensing, characterized in that, For executing the intelligent monitoring method for electrolyte fluid based on multimodal sensing as described in any one of claims 1 to 6, the system specifically includes the following logical functions: The parameter initialization and loading module is used to read the global parameter matrix and system configuration parameters obtained from offline calibration after power-on and write them into the static random access memory (SRAM). After the hardware cyclic redundancy check passes, the hardware interrupt sequence is enabled. The multimodal physical signal acquisition module is used to drive the multimodal physical sensing matrix to perform low-level signal synchronous acquisition, encapsulate the multi-source physical column vector sequence, and push it into the L1 cache through direct memory access controller DMA zero copy. The thermodynamic baseline compensation module is used to perform thermodynamic baseline compensation on the multi-source physical column vector sequence, generate pure multimodal features and write them into the compensation feature area of ​​SRAM, and flip the baseline compensation completion flag bit of the status register. The orthogonal decoupling and dimensionality reduction pooling module is used to concatenate and stitch pure multimodal features into a multimodal feature column vector, perform spatial orthogonal feature decoupling and high-dimensional feature scalar pooling compression mapping, output pure integrated electrical feature components and independent acoustic scalar feature components, and push them into the first-in-first-out (FIFO) buffer queue; The temporal high-order differential mapping module is used to extract temporal feature data from the FIFO buffer queue based on pointer addressing, perform temporal high-order differential mapping, extract the absolute dynamic trend energy value and write it into the multi-node evidence register group; The multi-node evidence fusion decision module is used to read the data in the multi-node evidence register group, substitute it into the dimensionless normalized exponential penalty model to calculate the basic probability score configuration confidence, call the DS evidence theory orthogonal and operator to eliminate node conflicts, solve the final decision confidence probability and latch it into the TrustZone security isolation area in real time. The integrated degradation assessment and adaptive control module is used to read the decision probability and dynamic energy in the register group within the isolation zone, perform engineering fusion mapping to generate the electrolyte fluid integrated degradation index, and perform hardware branch judgment on the electrolyte fluid integrated degradation index, the first warning threshold and the second protection threshold to drive the external execution terminal to execute the corresponding hierarchical interlocking protection action.

8. The intelligent monitoring system for electrolyte fluid based on multimodal sensing according to claim 7, characterized in that, The system includes a main control chip, which integrates the following functional units, including: The central processing unit (CPU) is used to coordinate data parsing, vector encapsulation, and system state machine scheduling. The floating-point unit (FPU) is used to respond to calls from the thermodynamic baseline compensation module and perform impedance spectrum temperature compensation calculations and ultrasonic attenuation temperature and pressure joint compensation calculations. The Tensor Accelerator (NPU) is used to respond to calls to the orthogonal decoupling and dimensionality reduction pooling modules, and to perform autocorrelation similarity matrix calculation, eigenvector decoupling, and Euclidean norm pooling compression mapping. The digital signal processing (DSP) core has a built-in digital differential filter, which is used to respond to the call of the time-series high-order differential mapping module and perform time-series feature data extraction and high-order differential mapping. The logic reasoning engine is used to respond to calls from the multi-node evidence fusion decision module, and to perform confidence calculations and orthogonal sum fusion operations.

9. The intelligent monitoring system for electrolyte fluid based on multimodal sensing according to claim 8, characterized in that, The system also includes a multimodal physical sensing matrix, memory, communication peripherals, and an external execution terminal, wherein: The multimodal physical sensing matrix includes a reference electrode-free miniature impedance measurement probe, an ultrasonic transceiver array, and a miniature integrated temperature and pressure sensor. The memory includes read-only memory (ROM) and static random access memory (SRAM); The main control chip also integrates a peripheral memory interface (FMC), a hardware CRC32 verification module, a direct memory access controller (DMA), a direct digital frequency synthesizer (DDS), an acoustic drive circuit, an advanced peripheral bus (APB), a status register, a receive register, a transmit holding register, a bus mapping unit, a multi-node evidence register group, and a TrustZone security isolation zone. External execution terminals include programmable logic controllers (PLCs), on-site audible and visual alarms, fluid circulation pumps, pipeline solenoid valves, and main circuit contactors.

Images