An automatic acquisition and analysis system of chemical experiment data

Abstract

The application relates to the technical field of data processing, and discloses an automatic acquisition and analysis system for chemical experiment data, which comprises an adaptive acquisition module, a time sequence alignment module, a phase state monitoring module and a compliance transfer module.The adaptive acquisition module is used for recording absolute time stamps based on the pixel change rate of a video stream and outputting visual data; the time sequence alignment module is used for extracting physical extreme points of high-frequency signals as constraint conditions, performing time sequence alignment on the visual data, and calculating a time axis distortion factor; the phase state monitoring module is used for constructing a multivariate partial derivative relationship matrix, adjusting a dynamic fault tolerance boundary according to the time axis distortion factor, and generating an abnormal early warning signal and a feedback instruction; and the compliance transfer module is used for constructing the aligned data and the early warning signal into a knowledge graph and performing hash encryption storage.The application solves the time sequence misalignment problem caused by asynchronous acquisition of multi-source heterogeneous data by using physical extreme point constraints, improves the reliability of abnormal detection by using a dynamic fault tolerance boundary, and establishes a tamper-proof experiment data traceability mechanism in combination with an encryption technology.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to an automatic acquisition and analysis system for chemical experimental data. Background Technology

[0002] In monitoring chemical experiments, it is typically necessary to utilize a combination of video equipment and various sensors to acquire visual images and physical parameters of the experimental site. With increasing automation in experiments, monitoring systems need to simultaneously process continuous video streams and high-frequency digital signals transmitted from sensors.

[0003] Because visual acquisition devices and physical sensors typically operate asynchronously, and their sampling frequencies differ significantly, in actual operation, network communication latency and clock deviations of individual hardware devices often lead to time-series misalignment between the low-frequency visual data received by the system and the high-frequency digital signal stream. Existing data processing methods primarily rely on coarse mapping based on system reception time, making it difficult to perform precise time-series alignment on such multi-source heterogeneous data, thus causing deviations in the correspondence between visual phenomena and physical parameters.

[0004] Based on the acquisition of misaligned data, conventional monitoring mechanisms mostly employ static thresholds for single physical variables to trigger out-of-range alarms. This approach severs the inherent covariant relationships between physical quantities in a chemical reaction, failing to identify abnormal states in the early stages when the proportions of multiple variables in the reaction system become unbalanced. Furthermore, the system generates additional computational errors when forcibly mapping inaccurately aligned data, easily triggering false alarms in dynamically fluctuating experimental environments. In addition, existing experimental monitoring systems mostly only save surface-level values ​​at the time of alarm when generating warning records. The warning results lack irreversible binding and encryption procedures between the underlying original data slices and the identification of the monitoring equipment, making the retained experimental data easily modifiable and failing to meet the requirements for data authenticity and integrity in chemical experiments during safety reviews and compliance traceability processes. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an automatic acquisition and analysis system for chemical experimental data, which solves the problems of time sequence misalignment of multi-source heterogeneous data, delayed detection of anomalies under multi-variable conditions, and imperfect anti-tampering and retention mechanisms for underlying verification data in existing chemical experimental data acquisition systems.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an automatic acquisition and analysis system for chemical experimental data, comprising: An adaptive acquisition module is used to receive a continuous video stream of a chemical experiment, record an absolute timestamp based on the pixel change rate of the continuous video stream to output low-frequency visual data, and receive feedback control commands. The timing alignment module is connected to the adaptive acquisition module and is used to receive the high-frequency digital signal stream and the low-frequency visual data, extract the physical extreme points of the high-frequency digital signal stream as constraints, perform timing alignment on the low-frequency visual data, and output the aligned data stream and time axis distortion factor. The phase monitoring module is connected to the timing alignment module, receives the alignment data stream and the time axis distortion factor to construct a multivariate partial derivative relationship matrix, generates an abnormal warning signal when the multivariate partial derivative relationship matrix exceeds the dynamic fault tolerance boundary adjusted by the time axis distortion factor, and sends the feedback control command to the adaptive acquisition module. The compliance flow module is connected to the phase monitoring module, receives the alignment data stream and the anomaly warning signal, constructs the alignment data stream and the anomaly warning signal into a knowledge graph and performs hash encryption storage.

[0007] Furthermore, the adaptive acquisition module extracts the image grayscale matrix of two adjacent video frames within the region of interest, calculates the absolute value of the difference between the image grayscale values ​​of the two adjacent video frames at corresponding coordinates, and determines the average displacement of the absolute value of the difference as the pixel change rate. This setting converts the jump in instrument data into a visual trigger signal by calculating the change in pixel intensity in a local area.

[0008] Furthermore, the adaptive acquisition module extracts the average pixel background grayscale value of the region of interest within a historical non-jumping period as a dynamic base, and superimposes a preset sensitivity bias coefficient onto the dynamic base to generate a dynamic threshold. When the pixel change rate is greater than the dynamic threshold and this state remains stable within a set continuous frame period, a hardware interrupt is triggered and absolute timestamp recording is executed. This setting, combined with background noise to generate the dynamic threshold and the introduction of continuous frame determination logic, reduces false triggers caused by ambient light flicker and improves the accuracy of timestamp binding.

[0009] Furthermore, upon receiving the feedback control command, the adaptive acquisition module narrows the boundary coordinate range of the region of interest in the continuous video stream and reduces the acquisition frequency of non-critical monitoring nodes based on the device dynamic activity mask in the feedback control command, allocating the freed-up computing resources to the optical character recognition and parsing process. This setting, through spatial coordinate clipping and local frequency reduction, frees up bus and computing resources, ensuring the visual parsing efficiency of the system under concurrent conditions.

[0010] Furthermore, the timing alignment module performs second-order difference operations on the high-frequency digital signal stream to obtain a changing acceleration sequence. When the absolute value of the changing acceleration sequence is greater than a preset steady-state threshold and is the only extreme peak within a local sliding time window, the corresponding moment is determined as the physical extreme point. This setting extracts the time nodes where the physical state of the reaction system undergoes abrupt changes as the benchmark for time axis alignment.

[0011] Furthermore, the temporal alignment module constructs rigid constraint boundaries using physical extreme points and restricts the computation path from crossing adjacent rigid constraint boundaries during dynamic time warping mapping. This projects discrete sampling points of low-frequency visual data onto corresponding coordinates on the high-frequency time axis. By comparing the time interval after mapping alignment with the original sampling time interval, the time axis distortion factor is extracted. This setup avoids cross-mapping between different chemical reaction stages of multi-source data and quantifies the temporal deviation during data alignment using the distortion factor.

[0012] Furthermore, the phase monitoring module uses the backward difference method to approximate the first-order partial derivatives of each physical parameter variable with respect to time in the aligned data stream, and calculates the instantaneous proportional relationship between the first-order partial derivatives of different physical parameter variables, thereby constructing a multivariate partial derivative relationship matrix. This setting establishes transient covariant relationships between different parameters, reflecting the proportional imbalance state in the energy and matter conversion process within the system.

[0013] Furthermore, the phase monitoring module calculates the penalty term using an exponential decay function of the natural logarithm base and tightens the preset basic fault tolerance boundary using a time axis distortion factor, generating a dynamic fault tolerance boundary that adaptively changes with the degree of data alignment distortion. This setting adjusts the anomaly detection boundary based on data errors caused by communication delays, preventing excessive data deviations from leading to missed detections.

[0014] Furthermore, the phase monitoring module calculates the norm distance between the multivariate partial derivative relationship matrix and the reference matrix corresponding to the current chemical experimental phase. When the norm distance exceeds the dynamic fault tolerance boundary and remains out of bounds for multiple consecutive sampling periods, an anomaly warning signal is triggered. This setting integrates spatial structure deviation and time dithering mechanism to perform anomaly determination, improving the reliability of the warning.

[0015] Furthermore, upon receiving an anomaly warning signal, the compliance workflow module extracts a standardized data slice within a preset time window and concatenates it with the absolute timestamp that triggered the warning, the current local knowledge graph structure, and the device's unique identifier using binary serialization. A secure hash algorithm is then used to extract the digital digest hash value, and the device's private key is used to perform an asymmetric encrypted signature. This setup establishes a tamper-proof association between the original collected data, diagnostic logic, and device identity, meeting the audit requirements for experimental data retention.

[0016] This invention provides an automated system for acquiring and analyzing chemical experimental data. It offers the following advantages: 1. This invention extracts the physical extreme points of the high-frequency digital signal stream as rigid constraints to perform constrained temporal alignment on low-frequency visual data acquired based on pixel changes in video frames, and outputs a time axis distortion factor. This method solves the problem of data timing misalignment caused by network latency or clock deviation in asynchronous operation environments of multi-source heterogeneous acquisition devices, and improves the accuracy of synchronous monitoring of vision and sensors in chemical experiments; 2. This invention constructs a multivariable partial derivative relationship matrix based on aligned data streams and adaptively tightens the dynamic fault-tolerant boundary using a time axis distortion factor and a decay function. This mechanism can identify imbalances in the covariant relationships between experimental parameters when a single physical parameter does not exceed its limit. Simultaneously, it uses the distortion factor to suppress computational noise caused by data alignment errors, achieving early and accurate warnings of abnormal states in chemical experiments. 3. When generating an anomaly warning signal, this invention concatenates and serializes the absolute timestamp triggering the warning, the local knowledge graph structure, and the device's unique identification code, and then uses a secure hash algorithm and asymmetric encryption technology to perform signature storage. This logic establishes a fixed record containing the warning node, underlying diagnostic data, and device identity, preventing subsequent tampering of experimental data and meeting the traceability and compliance auditing requirements for chemical experimental data retention. Attached Figure Description

[0017] Figure 1 This is a diagram of the overall system architecture of the present invention; Figure 2 This is a schematic diagram of the constrained spatiotemporal alignment principle based on dynamic anchor points of the present invention; Figure 3 This is a logic diagram of the dynamic degradation and closed-loop decision-making of the anomaly monitoring and judgment boundary in this invention; Figure 4 This is a schematic diagram of the multidimensional temporal knowledge graph topology and anomaly evidence splicing of the present invention; Figure 5 This is a comparison chart of the timing alignment accuracy of multi-source data in this invention; Figure 6 The following are comparison charts of the anomaly detection performance of the present invention. In the chart, (a) is a comparison chart of the early detection lead time of hidden anomalies, and (b) is a comparison chart of the system judgment accuracy.

[0018] Among them, 100 is the adaptive acquisition module; 200 is the timing alignment module; 300 is the phase monitoring module; and 400 is the compliance circulation module. Detailed Implementation

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

[0020] Please see the appendix Figure 1 -Appendix Figure 4 This invention provides an automatic acquisition and analysis system for chemical experimental data, comprising: The adaptive acquisition module 100 is used to trigger acquisition based on changes in video stream pixels and to perform hardware resource scheduling. The timing alignment module 200 is used to perform asynchronous data alignment based on dynamic anchor points and calculate distortion metrics. Phase monitoring module 300 is used to perform anomaly detection and feedback control based on multivariable partial derivative matrix and distortion factor; The Compliance Flow Module 400 is used to perform data standardization, knowledge graph construction, and compliant storage.

[0021] The adaptive acquisition module 100 runs within the edge computing node and is connected to the device monitoring camera and the timing alignment module 200. The adaptive acquisition module 100 receives the continuous video stream from the device monitoring camera, monitors the pixel change rate in the video stream, and when the pixel change rate exceeds the dynamic threshold, the adaptive acquisition module 100 triggers a hardware interrupt, records the system absolute timestamp, and pushes the corresponding video frame to the internal optical character recognition parsing queue. The adaptive acquisition module 100 then sends the parsed low-frequency visual data and the bound timestamp to the timing alignment module 200.

[0022] The timing alignment module 200 is connected to both the high-frequency digital sensor network and the phase monitoring module 300. The timing alignment module 200 synchronously receives the high-frequency digital signal stream acquired by the high-frequency digital sensor network, performs second-order difference operations on the high-frequency digital signal stream, and extracts physical extreme points as dynamic state anchor points. Using the dynamic state anchor points as constraints, the timing alignment module 200 performs nonlinear mapping alignment on the time axis of the low-frequency visual data and calculates the time axis distortion factor.

[0023] The phase monitoring module 300 is connected to the timing alignment module 200, the adaptive acquisition module 100, and the compliance flow module 400. The phase monitoring module 300 receives the aligned data stream and the time axis distortion factor, calculates the partial derivatives of each parameter variable in the data stream relative to time, constructs a multivariate partial derivative relationship matrix, compares the multivariate partial derivative relationship matrix with the preset phase constraint conditions, and introduces the time axis distortion factor to adjust the fault tolerance boundary of the anomaly judgment.

[0024] The phase monitoring module 300 generates an anomaly warning signal when the multivariable partial derivative relationship matrix exceeds the fault tolerance boundary. Simultaneously, it generates a feedback control command based on the currently identified system phase and sends it to the adaptive acquisition module 100. The adaptive acquisition module 100 receives the feedback control command and adjusts the optical character recognition calculation space range of the video frame and the polling acquisition frequency of the underlying device according to the command content.

[0025] The compliance flow module 400 receives structured data and early warning signals output by the phase monitoring module 300. The compliance flow module 400 constructs a relationship graph from the data nodes and performs data format standardization and hash encryption storage. The compliance flow module 400 provides an external configuration interface to receive user-defined phase constraint parameters and send them to the phase monitoring module 300.

[0026] The specific implementation principles of each module of the present invention will be explained in detail below with reference to the accompanying drawings.

[0027] In a specific embodiment, the adaptive acquisition module 100 runs on an edge computing node, realizing event-driven data acquisition and receiving downlink control commands to dynamically allocate underlying computing resources. To achieve the above functions, the adaptive acquisition module 100 specifically includes a video stream buffer unit, a pixel difference calculation unit, an interrupt flag unit, and a resource adaptive scheduling unit.

[0028] The video stream buffer unit receives continuous video streams from the monitoring camera. Edge computing nodes allocate a fixed-size circular buffer in memory for the continuous video stream. The video stream buffer unit manages video frame data using a first-in, first-out (FIFO) circular overwrite mechanism. When the circular buffer reaches its storage limit, newly written video frame data overwrites the oldest timestamped video frame data, thus maintaining a low-resource-consumption continuous video stream reading state. For the underlying memory addressing and pointer management of the circular buffer, those skilled in the art can use existing direct memory access techniques; the specific implementation of these techniques is well-known in the field and will not be elaborated upon here.

[0029] For traditional mechanical dials or digital displays, numerical jumps physically manifest as a sudden and drastic change in pixel grayscale values ​​within a specific local area. Based on this physical phenomenon, a pixel difference calculation unit is used to calculate the rate of pixel change within the region of interest (ROI) in adjacent video frames of a continuous video stream. The ROI refers to a local image region containing the fluctuating numerical values ​​or pointers of an analog dashboard, defined by a spatial coordinate matrix in the two-dimensional plane of the video frame. Let the current time be... The previous sampling time was The total number of pixels contained in the region of interest is denoted as To ensure the validity of matrix operations and prevent computational overflow caused by the denominator approaching zero, the system performs pre-verification. Is the pixel size greater than the set minimum valid pixel threshold? After confirming the region is valid, the pixel difference calculation unit extracts the image grayscale matrix of the region of interest from two adjacent frames and calculates the pixel change rate. The formula for calculating the pixel change rate is as follows: ; In the formula, express The pixel change rate calculated at each time step has the physical meaning of the average displacement of pixel intensity within the target area; This represents the set of spatial coordinates of the region of interest. Represents the two-dimensional spatial coordinates of a pixel within the region of interest; express At what coordinates is the video frame at that moment? The grayscale value of the image at that location; express Video frames at the same coordinates The grayscale value of the image at that location.

[0030] The interrupt flag unit is used to perform a timestamp binding action when the pixel change rate meets the trigger condition. In this embodiment, to adapt to the slow changes in laboratory ambient light, a dynamic threshold is used. Instead of being a fixed constant, the system extracts the average grayscale value of the pixel background within a historical period without jumps in the region of interest, using it as a dynamic base. A preset sensitivity bias coefficient is then superimposed on this base to generate the value, which ranges from 0 to 255 grayscale levels. To avoid false triggering by a single extreme value caused by ambient light flicker, the system introduces multi-dimensional judgment logic: when the pixel change rate... Greater than the dynamic threshold When this abrupt change remains stable within a set continuous microsecond-level frame period, it indicates a substantial jump in the value displayed on the analog dashboard. At this point, the interrupt flag unit sends a hardware-level interrupt request to the central processing unit (CPU) of the edge computing node. Upon responding to the interrupt request, the CPU immediately captures the absolute timestamp of the current system and extracts the current video frame data that triggered the interrupt from the circular buffer, binding it to the absolute timestamp in a header. After binding, the interrupt flag unit pushes the bound video frame to the internal optical character recognition (OCR) parsing queue. The temporal attribute of the parsed low-frequency visual data is determined by the physical instant of the data jump, thus eliminating data delay errors caused by the computation time of the OCR algorithm.

[0031] The resource adaptive scheduling unit receives feedback control commands from the phase monitoring module 300 and adjusts the acquisition parameters of the underlying hardware. The feedback control commands include the phase identifier of the current chemical reaction. When the feedback control command indicates that the current system is in a violent reaction phase, the resource adaptive scheduling unit performs a precise spatial coordinate system clipping operation. The resource adaptive scheduling unit reads the boundary coordinate vectors of the original region of interest, narrows the range of the coordinate vectors, and generates a new coordinate matrix after dimensionality reduction. Specifically, the resource adaptive scheduling unit clips the original complete dial area into a sub-region that only covers the area where the last digit of the value jumps, thereby reducing the total number of pixels in the region of interest. This reduces the matrix operation overhead of the pixel difference calculation unit.

[0032] While performing spatial coordinate clipping, the resource adaptive scheduling unit implements an edge computing power tilting strategy. The resource adaptive scheduling unit parses the device dynamic activity mask in the above instructions to identify non-critical monitoring nodes that do not participate in drastic energy exchange in the current phase. For these digital devices in a stable state, the resource adaptive scheduling unit modifies the polling cycle register parameters of their communication interfaces, reducing the data acquisition frequency of the corresponding digital devices and decreasing their occupation of system bus bandwidth. Based on the release of these resources, the resource adaptive scheduling unit allocates the released edge computing power and bus bandwidth to the parsing processes in the optical character recognition parsing queue by adjusting process priorities. This increases the allocation of local parsing computing resources for devices in drastically changing states, ensuring the real-time parsing performance of the system during high-frequency concurrent sudden changes.

[0033] In a specific embodiment, the timing alignment module 200 is used to receive multi-source heterogeneous data streams, perform spatiotemporal calibration according to the objective reaction dynamics law, and quantify the mathematical distortion caused by the alignment process. To achieve the above objectives, the timing alignment module 200 specifically includes an asynchronous data receiving unit, an anchor point extraction unit, a constrained regularization unit, and a distortion measurement unit.

[0034] The asynchronous data receiving unit is used to uniformly receive high-frequency digital signal streams and low-frequency visual analysis data streams bound with absolute timestamps. Due to the inherent system clock drift problem of different underlying acquisition devices, the timing of heterogeneous data streams upon arrival at the system is often disordered. To prevent memory overflow and ensure the real-time computation of subsequent algorithms, as a preferred approach, the asynchronous data receiving unit constructs a sliding time window structure based on a double-ended queue in the system memory. The high-frequency digital signal stream is continuously pushed into the queue by the digital sensor according to its internal high-frequency clock, while the low-frequency visual analysis data stream is non-periodically enqueued by the adaptive acquisition module 100 in a discrete, event-triggered manner. For the multiplexing and concurrent read / write control of the underlying network protocol stack of the aforementioned double-ended queue, those skilled in the art can use existing transmission control protocol concurrent processing technologies, the specific implementation of which is well-known in the field and will not be elaborated here.

[0035] The anchor point extraction unit performs second-order difference operations on the high-frequency digital signal stream to obtain the changing acceleration sequence. In chemical experimental scenarios, the second derivative of system temperature or pressure directly characterizes the fundamental change in the rate of heat absorption or release inside the reactor, and the extreme points of this physical quantity objectively map the transition boundaries of chemical kinetic phase states. Therefore, choosing the second-order difference of the high-frequency signal as the alignment reference has clear physical causality. Considering that the industrial environment easily introduces electromagnetic white noise into high-frequency sensors, the anchor point extraction unit performs smoothing preprocessing on the high-frequency digital signal stream using a moving average filtering algorithm before performing the difference operation.

[0036] To prevent distortion of the real signal caused by filtering, the sliding window size of this filtering algorithm is calculated based on the nominal sampling frequency of the corresponding sensor and the Nyquist sampling theorem. After smoothing, for any three consecutive valid sampling time points... , as well as The anchor point extraction unit calculates the high-frequency digital signal in The discrete second-order acceleration at time t. Its calculation formula is as follows: ; In the formula, express The acceleration of the reaction state change extracted at any time, the positive or negative value of which represents the increasing or decreasing trend of the reaction rate; The series represents the high-frequency digital signal values ​​acquired and filtered at the corresponding time.

[0037] To ensure the completeness of the complex division operation and prevent computational overflow caused by overlapping timestamps due to system lag, resulting in the denominator approaching zero, the anchor point extraction unit performs strict timing difference verification before inputting data. System Verification , as well as Are all three differences greater than the preset minimum hardware resolution constant (e.g., 10 times the system clock cycle)? -3 (on the order of seconds). If any check fails, the current sampled frame is determined to be invalid congested data and is discarded.

[0038] Based on the obtained acceleration sequence, the anchor point extraction unit further extracts local extreme points as reaction dynamics state anchor points. Since relying solely on a single extreme point is easily affected by transient physical disturbances, leading to distorted judgments, the anchor point extraction unit introduces multi-dimensional judgment logic. The system pre-calculates the noise floor variance under the stable no-load condition of the equipment and uses a preset multiple of this noise floor variance as the steady-state threshold. When the acceleration... The absolute value is greater than the steady-state threshold, and the In When the local sliding time window centered on the peak point is the only extreme peak, the anchor point extraction unit will... Identify it as a dynamic state anchor point and push it into the anchor point set. The extracted anchor points are not only mathematical nodes on the timeline, but also hard physical boundaries for subsequent alignment of multi-source data conditions.

[0039] Constrained warping units are used to eliminate spatiotemporal misalignments between heterogeneous devices. Conventional interpolation methods rely solely on mathematical timeline point filling, often neglecting the physical synchronicity of changes in multiple physical parameters. Based on previously extracted physical features, constrained warping units use a set of anchor points... The time nodes in the time structure serve as rigid constraint boundaries, performing nonlinear mapping and alignment on the time axis of low-frequency visual data. Specifically, the constrained warping unit divides the high-frequency signal time axis and the low-frequency data time axis into multiple sub-intervals isolated by dynamic state anchor points. When performing the dynamic time warping algorithm to search for the optimal alignment path, the constrained warping unit forcibly restricts the computation space of the cumulative distance matrix from crossing the intervals formed by adjacent anchor points. Guided by this boundary isolation, the constrained warping unit projects the discrete sampling points of low-frequency visual data onto the corresponding time coordinates of the high-frequency time axis, thereby achieving strict alignment of physical phase transition features in the same time domain dimension and preventing illegal mapping between heterogeneous data and different chemical reaction stages.

[0040] The distortion metric unit is used to quantify the degree to which the aforementioned nonlinear mapping process distorts the objectivity of physical time passage. During the execution of constrained dynamic time warping mapping, the original low-frequency data timeline inevitably undergoes local stretching or compression transformations. For two adjacent low-frequency visual data sampling points, let the absolute timestamp difference bound in the original receiving queue be denoted as... After mapping and alignment, the two sampling points generate a new time difference on a unified high-frequency time axis, let's call it... The distortion measurement unit calculates the duration distortion factor by comparing the difference between the two, and the calculation formula is as follows: ; In the formula, This represents the time axis distortion factor calculated within the current time slice. It uses absolute value operations to ensure that the distortion measure is a non-negative penalty term regardless of whether the time axis is stretched or compressed. Indicates the original sampling time interval; Indicates the time interval after mapping alignment; This represents the system's preset minimum positive bias constant. In this embodiment, The preferred value range is 10. -6 Up to 10 -8 Seconds. Introduce a positive bias constant. The technical purpose is to address extreme hardware failures, such as severe frame drops and retransmissions from the underlying camera, that could lead to... When the value approaches zero, the validity of the denominator is guaranteed and operational overflow is prevented. The value of this factor directly reflects the confidence level of the current data. The larger the value of this factor, the more severe the underlying asynchronous communication latency, and the higher the mathematical distortion required to achieve data feature alignment, thus providing a basic quantitative basis for the adaptive adjustment of the fault tolerance boundary for subsequent anomaly detection.

[0041] In a specific embodiment, the phase monitoring module 300 is used to extract the dynamic covariance relationship between multiple variables for state diagnosis, and dynamically adjust the judgment weights in conjunction with the distortion factor aligned with the underlying data. To achieve the above technical objectives, the phase monitoring module 300 specifically includes a phase partitioning unit, a matrix calculation unit, a fault-tolerant boundary degradation unit, and a closed-loop decision unit.

[0042] Phase partitioning units are used to define the discrete phases of the entire chemical reaction process and construct the system state vector. In multi-parameter coupled chemical reaction systems, different reaction stages, such as the heating-feeding period and the isothermal catalytic period, possess specific physical and thermodynamic constraints. This is based on the set of kinetic state anchor points extracted by the aforementioned time-alignment module. The phase partitioning unit divides the complete chemical process flow into multiple discrete phases along the time axis and assigns phase index numbers. ,in It is a positive integer.

[0043] Simultaneously, the phase partitioning unit receives the spatiotemporal calibration data stream output by the timing alignment module 200 and extracts it from the data. Synchronized physical parameters to construct the current moment System multivariable state vector The mathematical expression for this state vector is: ,in Indicates the first A physical parameter in The alignment values ​​at different times, these parameters cover high-frequency continuous signals such as reactor temperature and pressure, as well as low-frequency discrete visual signals such as liquid level scale.

[0044] The matrix computation unit is used to calculate the time partial derivatives of parameter variables and construct a multivariate partial derivative relationship matrix. The absolute threshold of a single physical parameter is insufficient to expose hidden precursors of runaway reactions, while the relative rate of change between parameters can intuitively reflect the internal energy and matter conversion ratio of the system. For the discrete data stream received by the system, the matrix computation unit uses the backward difference method to approximate the first-order partial derivatives of each parameter variable relative to time, i.e. ,in This represents the aligned discrete time step. Subsequently, the matrix computation unit constructs a matrix with dimension... Multivariable partial derivative relationship matrix Elements inside the matrix The calculation formula is as follows: ; In the formula, Indicates in At that moment, the The rate of change of the first physical parameter and the first The instantaneous proportional relationship of the rate of change of a physical parameter has the physical meaning of the sensitivity mapping of different physical quantities in the energy interaction process when the system is in the current phase state. and These represent the state vectors of the first and second digits, respectively. The and the first To ensure the completeness of matrix operations and prevent division-by-zero overflow caused by a physical parameter changing at a rate of zero or at a critical negative value under steady-state conditions, the system introduces a sign protection function in the underlying calculation logic of the denominator term. That is, the denominator is actually represented as... (The sign function is forced to be positive when the derivative is 0), and a minimal nonzero constant is introduced into the denominator. As a preferred method, The value is determined based on the accuracy and noise floor lower limit of the specific underlying sensor, and is usually set to 10. -5 Up to 10 -4 Magnitude.

[0045] The fault-tolerant boundary degradation unit is used to adaptively tighten the anomaly detection boundary using the underlying time axis distortion factor. During data fusion from heterogeneous devices, underlying communication delays and clock drift can cause mathematical distortions in the aligned data sequence. Assigning excessively high confidence levels to highly distorted data can easily lead to missed anomaly detections. To address this issue, the fault-tolerant boundary degradation unit acquires the time axis distortion factor synchronously output by the timing alignment module 200. This is then directly incorporated into the calculation logic of the fault tolerance boundary. The calculation formula for the dynamic fault tolerance boundary is as follows: ; In the formula, This represents the current dynamic fault tolerance boundary after distortion factor correction; This represents the system's preset basic fault tolerance boundary constant, the value of which is determined based on the historical normal dispersion of the no-load calibration experiment. This represents the time axis distortion factor within the current time slice; This represents the penalty attenuation coefficient. In this embodiment, The preferred value range is 0.5 to 2.0, and its specific value is determined by the average packet arrival delay jitter rate of the field industrial communication network. The purpose of introducing the exponential decay function is to utilize the nonlinear smoothing characteristics of the natural logarithm base to avoid false alarms caused by a sudden drop in the judgment boundary due to small fluctuations in the distortion factor. At the same time, it rapidly tightens the boundary when the alignment is severely distorted, enabling the system to adopt a more conservative and safe judgment strategy.

[0046] The closed-loop decision unit performs spatial feature comparison and generates feedback control commands downstream. The closed-loop decision unit stores various discrete phase states. Corresponding preset dynamic constraints Mathematically, this constraint is represented as a baseline relation matrix under ideal conditions. The closed-loop decision unit calculates the currently generated multivariate partial derivative relation matrix. The Frobenius norm distance between the current actual operating condition and the corresponding phase reference matrix is ​​used to quantify the global spatial deviation of the current actual operating condition from the ideal reaction dynamic trajectory.

[0047] The system will compare the deviation with the dynamic fault tolerance boundary. Continuous comparison is performed. To avoid control oscillations caused by single-point data noise, the closed-loop decision unit executes anti-shake weighted logic based on a time window: when the calculated deviation exceeds the dynamic fault tolerance boundary... And this out-of-bounds state is continuous Each sampling period (e.g.) If the condition remains unchanged within a given range, it indicates that the covariant balance between parameters has been substantially broken. At this point, the closed-loop decision unit immediately generates an anomaly warning signal.

[0048] At the same time, the closed-loop decision-making unit is based on the currently identified phase state. Generate feedback control commands carrying computing power reallocation parameters. The closed-loop decision-making unit outputs the early warning signal to the compliance flow module 400 and sends feedback control instructions. The resource adaptive scheduling unit, which is sent to the adaptive acquisition module, triggers the tilting of the front-end acquisition frequency and the clipping of the optical character recognition calculation space range, thereby constructing a hardware-level closed-loop system covering physical perception, algorithm judgment and edge control.

[0049] In a specific embodiment, the compliance transfer module 400 is used to perform structured modeling of the large-scale heterogeneous verification data generated by the underlying closed loop and establish an anti-tampering experimental traceability system to meet the data compliance audit requirements of the industrial and scientific research fields. To achieve the above technical objectives, the compliance transfer module 400 specifically includes a configuration and standardization unit, a feature metadata extraction unit, a knowledge graph construction unit, and an anomaly evidence and log unit.

[0050] To overcome the physical barriers of inconsistent data dimensions and varying precision from multiple underlying devices, a configuration and standardization unit provides a user-facing interface and performs underlying data cleaning for heterogeneous data. For the specific process requirements of different chemical experiments, the system offers a visual configuration interface that receives user-defined phase constraint parameters. Based on the differences in physical specifications among multi-source sensors, the configuration and standardization unit performs unit dimension alignment and significant digit reduction on the parallel-input heterogeneous data streams for format standardization. The system uses a built-in SI conversion dictionary to uniformly map raw acquired values ​​of different dimensions to a standard physical unit space. For any physical parameter's raw sampled value... Its linear standardization conversion formula is as follows: ; In the formula, Indicates in The physical parameter values ​​after format standardization; This represents the dimension conversion ratio for the sensor to which this physical parameter belongs; This represents the zero-point offset constant of the system. As a preferred method, and The values ​​are directly derived from the slope and intercept parameters in the factory calibration description files of each underlying sensor. After obtaining data with consistent dimensions, to avoid calculation errors introduced by the algorithm due to handling invalid mantissas, the configuration and normalization unit further performs significant digit truncation. The system queries the nominal physical resolution of the corresponding sensor (e.g., 0.01℃ for a temperature sensor) and forces the truncation to be... The floating-point precision is aligned to the least significant bit corresponding to the resolution, eliminating invalid high-order floating-point noise that exceeds the hardware's perception limit, thereby reducing the overall storage overhead of the system from the source.

[0051] After basic data cleaning, the feature metadata extraction unit extracts core business nodes with traceability value from the massive data of the preceding chain. In long-cycle chemical reactions, storing the entire high-frequency video and signal streams can easily overload the storage medium. Based on event-driven logic, the feature metadata extraction unit extracts five types of core data as metadata nodes from the preceding processing chain: first, hardware-level absolute timestamps, used to anchor the objective physical moment of abnormal data occurrence; second, a set of dynamic state anchor points, used to characterize the spatiotemporal coordinates of sudden changes in the heat absorption and release rates of the reaction system; third, time axis distortion factors, used to quantify the confidence level of mathematical distortion when aligning heterogeneous data; fourth, a multivariate partial derivative relationship matrix, used to record the transient covariance characteristics between system parameters; and fifth, anomaly warning signals, used to solidify the trigger point evidence for system intervention decisions.

[0052] Based on the extracted discrete metadata, the knowledge graph construction unit builds a multi-dimensional time-series knowledge graph of chemical experiment batches. Conventional relational databases struggle to intuitively represent the complex coupling relationships of multiple parameters across time and causality. Therefore, the knowledge graph construction unit defines the graph data structure. ,in This is a collection of entity nodes containing the above five types of core metadata. It is the set of directed edges connecting nodes.

[0053] The knowledge graph construction unit establishes directed edges with specific temporal semantics between absolute timestamp nodes and dynamic state anchor points, and simultaneously establishes causal semantic connections between the partial derivative relation matrix and anomaly warning signals. Through this structured association process, the system transforms a single experimental data pipeline into a three-dimensional network knowledge graph encompassing temporal evolution, phase transitions, and anomaly determination logic. This allows post-event personnel to avoid tedious log retrieval and quickly trace the physical root of any anomalous event along the graph's topological relationships.

[0054] When an actual safety breach occurs on-site, the anomaly evidence preservation and logging unit is used to extract hash digests from data slices and retain them immutably. Because the experimental data at this time has extremely high value for incident identification and compliance review, the anomaly evidence preservation and logging unit immediately extracts standardized data slices centered on the warning time and extending forward and backward by preset time windows (e.g., 30 seconds before and after). The system then concatenates this data slice, the current knowledge graph subgraph, and the generated standardized experimental report using binary serialization, and extracts the data digest using a secure hash algorithm. Its one-way hash operation structure is as follows: ; In the formula, This represents a fixed-length digital digest hash value generated for the current anomalous event; Indicates the selected secure hashing algorithm; This represents a sequence of extracted, multidimensionally normalized data slices. This represents the local map structure data when an anomaly occurs; Indicates the absolute timestamp that triggered the alert; This represents the physically unique identifier (such as a MAC address or trusted execution environment hardware serial number) of the edge computing node or core controller that generated the warning signal; symbol This represents the sequential concatenation operation of the aforementioned binary data blocks. The technical purpose of this hash operation is to extract a globally unique digital fingerprint for the physical state and hardware identity of the entire system at an abnormal instant, so that any slight tampering with the original underlying data will cause the calculated hash value to change drastically.

[0055] After completing the digital fingerprint extraction, to ensure the non-repudiation of the data source, the anomaly verification and log unit uses the device-specific private key stored in the hardware security module to... Asymmetric encryption digital signature operations are performed. Finally, the anomaly verification and logging unit stores the signed hash fingerprint in the backend immutable log system, while the original massive payload data slices are archived in a conventional distributed file system. Through this engineering implementation mechanism that physically isolates the original data payload from the compliant tamper-proof fingerprint, the system ensures full lifecycle data legality auditing and identity traceability capabilities while avoiding network bandwidth and consensus performance bottlenecks caused by directly writing large-scale industrial time-series data into the distributed ledger.

[0056] To enable those skilled in the art to better understand the present invention, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This embodiment uses high-pressure exothermic polymerization reaction monitoring as its application scenario.

[0057] In the initial stage of the reaction (heating phase), the system is in a steady state. The adaptive acquisition module 100 detects no bubbles and no color change within the viewing mirror area, and the pixel change rate... Maintain at a low level (below the dynamic threshold) At this time, the module controls the underlying polling frequency to maintain a low-power mode of 1Hz.

[0058] When the polymerization reaction enters its critical phase, the camera captures tiny bubbles churning on the liquid surface inside the viewing mirror, causing a change in pixel rate. The dynamic threshold is instantly exceeded. The adaptive acquisition module 100 immediately triggers a hardware interrupt, locks the current microsecond-level absolute timestamp, and automatically clips the region of interest (ROI) to a localized area containing only the liquid level scale line and the dense bubble area. At the same time, the system increases the polling frequency of the pressure and temperature sensors to 100Hz by issuing a command, prioritizing the computing power for the optical character recognition (OCR) process to parse the liquid level gauge reading.

[0059] Due to network fluctuations, the visual resolution data (liquid level height) arrived at the server approximately 150ms later than the temperature sensor data. Upon receiving the data stream, the timing alignment module 200 extracted the extreme point of the second derivative of the temperature sensor data (i.e., the moment of maximum heat release rate) as a kinetic anchor point. Using this anchor point as a reference, the module employed a constrained dynamic time warping algorithm to forcibly stretch and align the lagging visual liquid level data to the time axis of the temperature abrupt change, and calculated the time axis distortion at this point. The value is 0.15 (indicating a certain degree of timing distortion).

[0060] The phase monitoring module 300 constructs a state vector containing [temperature, pressure, liquid level] and calculates a multivariable partial derivative relationship matrix.

[0061] Under normal operating conditions: Based on the preset ideal model, the ratio of the rate of change of pressure with increasing temperature It should be kept near the linear constant.

[0062] Abnormal operating condition simulation: In this implementation, polymer buildup inside the reactor caused localized overheating. Although individual temperature values ​​did not exceed the limit, the rate of temperature rise relative to the rate of pressure change deviated abnormally. At this point, the partial derivative relationship matrix... The corresponding element in [the data] undergoes a mutation. This is due to the distortion factor calculated above. The system uses formulas The fault tolerance boundary was automatically tightened.

[0063] The calculated spatial characteristic deviation quickly exceeded the tightened fault tolerance boundary. The system immediately generated a phase imbalance warning and issued a feedback command to force the adaptive acquisition module to lock the full video stream recording at 100%, while simultaneously sending an emergency cooling signal to the PLC control system.

[0064] After the alert is triggered, the compliance workflow module 400 extracts a 30-second data slice before and after the alert, and extracts the absolute timestamp, anchor point set, and anomaly matrix. The system then generates a hash digest. The device's private key is then used to digitally sign the data. This signature data is written in real time to an immutable audit log, forming a complete chain of evidence that includes visual evidence (bubble bubbling) and data evidence (partial derivative imbalance).

[0065] Comparison group settings: Control group: The traditional timed polling (fixed 10Hz) was used for data acquisition, and the static threshold alarm was based on a single variable (temperature > 180℃), without time alignment and partial derivative matrix analysis.

[0066] Experimental group (this invention): Adaptive acquisition (dynamic adjustment from 1-100Hz), constrained regularization alignment based on dynamic anchor points, and monitoring of multivariable partial derivative matrix.

[0067] The experimental data are shown in the table below: Table 1 Comparison of Temporal Alignment Accuracy of Multi-Source Data

[0068] Summarize: According to Table 1 and Appendix Figure 5 As a result, under the network latency and jitter environment in industrial settings, the data alignment error of traditional timed polling schemes shows a significant worsening trend, failing to meet the timeliness requirements of multi-parameter joint analysis. This invention introduces physical dynamic anchor points as rigid constraint boundaries, strictly controlling the alignment error of visual and sensor data under multi-network fluctuation conditions to within 25ms (less than the frame period of a single visual sampling). The alignment accuracy is improved by 88%-92%, effectively overcoming the asynchronous misalignment problem of data features caused by clock drift and network congestion in multi-source devices, providing a high-confidence data foundation for subsequent joint diagnosis.

[0069] Table 2 Comparison of Anomaly Detection Performance

[0070] Summarize: According to Table 2 and Appendix Figure 6 As a result, traditional single static threshold monitoring can only trigger passive alarms at the critical stage of an accident (such as when the temperature exceeds the set limit). This invention, by constructing a multivariate partial derivative relationship matrix, can effectively identify early, hidden precursors where a single parameter is not exceeded but the energy conversion ratio within the system has become unbalanced, achieving an effective early warning with an average advance warning time of 42.5 seconds. Furthermore, the system incorporates a time axis distortion factor as a penalty term into the calculation logic of the dynamic fault-tolerant boundary, effectively filtering out numerical calculation noise caused by high-frequency data alignment. While ensuring the sensitivity of early warning, it reduces the false alarm rate to 0.3% and the false alarm rate to 1.2%.

[0071] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An automatic acquisition and analysis system for chemical experiment data, characterized in that, include: An adaptive acquisition module is used to receive a continuous video stream of a chemical experiment, record an absolute timestamp based on the pixel change rate of the continuous video stream to output low-frequency visual data, and receive feedback control commands. The timing alignment module is connected to the adaptive acquisition module and is used to receive the high-frequency digital signal stream and the low-frequency visual data, extract the physical extreme points of the high-frequency digital signal stream as constraints, perform timing alignment on the low-frequency visual data, and output the aligned data stream and time axis distortion factor. The phase monitoring module is connected to the timing alignment module, receives the alignment data stream and the time axis distortion factor to construct a multivariate partial derivative relationship matrix, generates an abnormal warning signal when the multivariate partial derivative relationship matrix exceeds the dynamic fault tolerance boundary adjusted by the time axis distortion factor, and sends the feedback control command to the adaptive acquisition module. The compliance flow module is connected to the phase monitoring module, receives the alignment data stream and the anomaly warning signal, constructs the alignment data stream and the anomaly warning signal into a knowledge graph and performs hash encryption storage.

2. The automatic acquisition and analysis system for chemical experimental data according to claim 1, wherein, The adaptive acquisition module extracts the image grayscale matrix of two adjacent video frames within the region of interest, calculates the absolute value of the difference between the image grayscale values ​​of the two adjacent video frames at corresponding coordinates, and determines the average displacement of the absolute value of the difference as the pixel change rate.

3. The system according to claim 2, wherein the system is configured to automatically collect and analyze the data from the chemical experiment. The adaptive acquisition module extracts the average grayscale value of the pixel background in the region of interest within a historical period without jumps as a dynamic base, and superimposes a preset sensitivity bias coefficient on the dynamic base to generate a dynamic threshold. When the pixel change rate is greater than the dynamic threshold and the state remains stable within a set continuous frame period, a hardware interrupt is triggered and the recording of the absolute timestamp is executed.

4. The automatic acquisition and analysis system for chemical experimental data according to claim 1, characterized in that, After receiving the feedback control command, the adaptive acquisition module narrows the boundary coordinate range of the region of interest in the continuous video stream, and reduces the acquisition frequency of non-critical monitoring nodes according to the device dynamic activity mask in the feedback control command, and allocates the released computing resources to the optical character recognition and parsing process.

5. The automatic acquisition and analysis system for chemical experimental data according to claim 1, characterized in that, The timing alignment module performs a second-order difference operation on the high-frequency digital signal stream to obtain a changing acceleration sequence. When the absolute value of the changing acceleration sequence is greater than a preset steady-state threshold and is the only extreme peak point within a local sliding time window, the corresponding time is determined as the physical extreme point.

6. The automatic acquisition and analysis system for chemical experimental data according to claim 5, characterized in that, The time alignment module constructs rigid constraint boundaries using the physical extreme points and restricts the calculation path from crossing adjacent rigid constraint boundaries when performing dynamic time warping mapping. It projects the discrete sampling points of the low-frequency visual data onto the corresponding coordinates of the high-frequency time axis and extracts the time axis distortion factor by comparing the difference between the time interval after mapping alignment and the original sampling time interval.

7. The automatic acquisition and analysis system for chemical experimental data according to claim 1, characterized in that, The phase monitoring module uses the backward difference method to approximate the first-order partial derivatives of each physical parameter variable in the aligned data stream relative to time, and calculates the instantaneous proportional relationship between the first-order partial derivatives of different physical parameter variables, thereby constructing the multivariate partial derivative relationship matrix.

8. The automatic acquisition and analysis system for chemical experimental data according to claim 7, characterized in that, The phase monitoring module calculates the penalty term using the exponential decay function of the natural logarithm base, and uses the time axis distortion factor to perform a tightening operation on the preset basic fault tolerance boundary, generating the dynamic fault tolerance boundary that adaptively changes with the degree of data alignment distortion.

9. The automatic acquisition and analysis system for chemical experimental data according to claim 8, characterized in that, The phase monitoring module calculates the norm distance between the multivariable partial derivative relationship matrix and the reference matrix corresponding to the current chemical experimental phase. When the norm distance is greater than the dynamic fault tolerance boundary and remains out of bounds for multiple consecutive sampling periods, the abnormal warning signal is triggered.

10. The automatic acquisition and analysis system for chemical experimental data according to claim 1, characterized in that, When the compliance flow module receives the abnormal warning signal, it extracts a standardized data slice within a preset time window and concatenates it with the absolute timestamp that triggered the warning, the current local knowledge graph structure, and the device's unique identification code in binary serialization. It then uses a secure hash algorithm to extract the digital digest hash value and performs asymmetric encryption signature using the device's private key.

Images