An enterprise production data traceability management method based on artificial intelligence

Abstract

The application relates to the technical field of intelligent manufacturing, and discloses an enterprise production data traceability management and control method based on artificial intelligence. The present application implants a digital coding disturbance sequence as a physical tracking fingerprint, combines kinematics modeling based on the Lagrange perspective, and solves the technical pain point that precise tracing cannot be achieved in a non-steady production environment. Unlike traditional statistical matching methods that only rely on waveform geometric similarity, the present application constructs a physical perception time sequence alignment mechanism, and forcibly requires the time mapping path to follow the displacement conservation and entropy increase law. Furthermore, the present application uses mass balance verification to physically check the tracing results. This technical leap from signal correlation to physical causality effectively eliminates false alignment and quantization errors caused by shutdown and speed changes, ensures the absolute accuracy and non-tamperability of the responsibility entity lock, and greatly improves the robustness and legal credibility of the industrial data management and control system.

Description

Technical Field

[0001] This application relates to the field of intelligent manufacturing technology, and in particular to an artificial intelligence-based method for tracing and managing enterprise production data. Background Technology

[0002] In modern process-oriented industrial manufacturing, particularly in production scenarios such as lithium-ion battery electrode preparation, chemical fiber spinning, and continuous casting and rolling of steel, the production process exhibits significant long-flow and continuous-flow characteristics. Such production lines typically consist of dozens or even hundreds of processes connected in series. Once raw materials are input into the production line, they undergo physical or chemical processing steps in the form of fluids, powders, or continuous rolls, including mixing, coating, drying, rolling, slitting, or melting, refining, and continuous casting. To achieve digital control of the production process, companies typically deploy high-frequency sensors at various key processes to collect real-time process parameters such as temperature, pressure, tension, operating speed, voltage, and current. This massive amount of time-series data constitutes the fundamental data source for companies to monitor production quality, predict equipment failures, and optimize process parameters. The aim is to achieve full lifecycle traceability and control from raw materials to finished products through the analysis of the entire process data.

[0003] However, in existing production data traceability and correlation analysis technologies, data alignment for the aforementioned long-process continuous production scenarios mainly relies on estimation methods based on fixed timestamp offsets or logical deduction methods based on first-in-first-out queues. This traditional method is based on an idealized assumption: that the flow rate of materials between processes is constant, or that the residence time of materials in buffer zones is strictly controllable. However, in the complex industrial production environment, this assumption is often difficult to uphold. Due to various physical factors such as fluctuations in power supply voltage, wear and slippage of mechanical transmission components, changes in conveyor belt tension, and differences in the rheological properties of different batches of materials, the actual operating speed of the production line will exhibit nonlinear dynamic fluctuations. Furthermore, to balance the production cycle time of upstream and downstream processes, production lines often include buffer tanks or accumulation areas with variable capacities; the actual residence time of materials in these areas has extremely high uncertainty and randomness.

[0004] This nonlinear fluctuation in the physical world directly leads to temporal deviations in the data world. When using traditional fixed-time models to correlate process parameters collected from upstream processes with quality data detected in downstream processes, significant timeline misalignment occurs. Since fluids or continuous roll surfaces are difficult to physically trace like discrete workpieces through marking or attaching electronic tags, data timestamps become the sole basis for correlation. Once cumulative errors occur in timeline estimation, it leads to a severe mismatch between feature data and tag data. For example, the system might incorrectly attribute the cause of a current product defect to fluctuations in process parameters unrelated to the product that occurred minutes or even hours ago. This nonlinear drift and alignment bias in spatiotemporal data not only severely interferes with the accurate learning of the causal relationship between production processes and product quality by artificial intelligence models, causing model training failure, but also prevents enterprises from achieving accurate anomaly interception and real-time feedback during production, resulting in significant resource waste and control blind spots. Summary of the Invention

[0005] This application proposes an artificial intelligence-based method for tracing and managing enterprise production data to address the problems mentioned in the background art.

[0006] To achieve the above objectives, this application adopts the following technical solution: a method for traceability and control of enterprise production data based on artificial intelligence, comprising the following steps:

[0007] Step S1: The controller obtains the basic process setting parameters and process safety constraint range of the current production process. The controller calculates the modulation depth threshold according to the process safety constraint range. The controller generates a digitally encoded disturbance sequence that meets the modulation depth threshold limit. The controller superimposes the digitally encoded disturbance sequence onto the basic process setting parameters, generates a physical tracer control command, and sends it to the production execution mechanism. The production execution mechanism drives material transfer according to the physical tracer control command.

[0008] Step S2: The speed sensor collects the instantaneous transmission speed sequence on the material transmission link in real time. The processor performs time integration on the instantaneous transmission speed sequence to calculate the cumulative physical displacement value. Based on the cumulative physical displacement value and the preset physical link design length, the processor calculates the earliest and latest arrival times of the material to the downstream detection point and constructs an effective kinematic time window.

[0009] Step S3: The downstream sensor collects the downstream response data sequence. The processor uses the kinematic effective time window as the search space boundary and uses the physical constraint dynamic time warping algorithm to calculate the optimal time mapping path between the digitally encoded perturbation sequence and the downstream response data sequence. The physical constraint dynamic time warping algorithm includes a velocity consistency penalty term, which is determined by the difference between the instantaneous transmission velocity sequence and the virtual flow velocity ratio. The virtual flow velocity ratio is the local slope of the optimal time mapping path.

[0010] Step S4: The processor uses the optimal time mapping path to map the data points in the downstream response data sequence back to the absolute physical input time. The processor obtains the material input quality flow rate corresponding to the absolute physical input time and performs a quality balance check. After passing the quality balance check, the processor retrieves the responsible entity identifier in the production log based on the absolute physical input time and generates a traceability control data vector containing the responsible entity identifier.

[0011] Furthermore, in step S1, the controller calculates the modulation depth threshold based on the process safety constraints, specifically including:

[0012] The controller reads the pre-stored standard deviation of the sensor's background noise, and calculates the first physical distance between the basic process setting parameters and the upper limit of process safety in the process safety constraint range, and the second physical distance between the basic process setting parameters and the lower limit of process safety in the process safety constraint range.

[0013] The controller defines the smaller of the first physical distance and the second physical distance as the minimum safe distance, and compares the minimum safe distance with three times the standard deviation of the sensor's background noise.

[0014] When the minimum safe distance is less than three times the standard deviation of the sensor's noise floor, the controller sets the modulation depth threshold value directly to zero. When the minimum safe distance is greater than or equal to three times the standard deviation of the sensor's noise floor, the controller calls the asymmetric safety barrier model to calculate the modulation depth threshold.

[0015] When the controller invokes the asymmetric security barrier model, it obtains the risk sensitivity length parameter, calculates the negative of the ratio of the minimum safe distance to the risk sensitivity length parameter, performs natural exponential operation on the negative of the ratio to obtain the exponential decay term, calculates the value minus the exponential decay term to obtain the security convergence factor, calculates the product of the minimum safe distance, the security convergence factor and the preset security margin coefficient, and assigns the result of the product to the modulation depth threshold to complete the calculation of the physical observability constraint of the modulation depth threshold.

[0016] Further, in step S1, the controller generates a digitally coded perturbation sequence that meets the modulation depth threshold limit, and superimposes the digitally coded perturbation sequence onto the basic process setting parameters to generate a physical tracer control command, specifically including:

[0017] The controller reads the step response time constant of the production actuator, and multiplies the step response time constant by the bandwidth redundancy coefficient to obtain the minimum symbol holding time. The value of the bandwidth redundancy coefficient is not less than three.

[0018] The controller runs a discrete chaotic mapping algorithm to generate an original chaotic numerical sequence. The controller performs a sample-and-hold operation on the original chaotic numerical sequence based on the minimum symbol holding time. The controller performs symbolic binary processing on the numerical sequence after the sample-and-hold operation to generate a bandwidth-limited digital coded perturbation sequence.

[0019] The controller calculates the product of the digitally encoded disturbance sequence and the modulation depth threshold to obtain the theoretical disturbance amplitude sequence. The controller performs an addition operation on the theoretical disturbance amplitude sequence and the basic process setting parameters to generate the original superimposed signal. The controller uses a dynamic smooth convolution kernel to perform time-domain convolution filtering on the original superimposed signal. The maximum value of the second derivative of the dynamic smooth convolution kernel is less than the maximum allowable acceleration limit of the production actuator.

[0020] The controller eliminates the step abruption edges in the original superimposed signal through time-domain convolution filtering, generating a continuous and differentiable smooth curve signal. The controller outputs the smooth curve signal as a physical tracer control command.

[0021] Furthermore, in step S2, the speed sensor collects the instantaneous transmission speed sequence on the material transport link in real time, and the processor performs time integration on the instantaneous transmission speed sequence to calculate the cumulative physical displacement value, specifically including:

[0022] The processor obtains the material rheological characteristic identifier of the current production batch from the manufacturing execution system. Based on the material rheological characteristic identifier, the processor queries the preset flow pattern distribution coefficient database and obtains the rheological correction factor that matches the current operating conditions.

[0023] The processor sets the zero-speed dead zone threshold of the sensor. The processor performs zero-speed dead zone logic judgment on the raw speed readings collected by the speed sensor. When the raw speed reading is lower than the zero-speed dead zone threshold of the sensor, the processor determines that the effective material transport speed at the current moment is zero. When the raw speed reading is higher than the zero-speed dead zone threshold of the sensor, the processor calculates the product of the raw speed reading and the rheological correction factor to obtain the physically calibrated effective material transport speed.

[0024] The processor uses an anti-drift discrete Riemann sum algorithm to perform time-domain accumulation operation on the effective material transport speed. In each system sampling period, the processor calculates the product of the effective material transport speed and the system sampling period, and adds the product to the historical accumulated value. The processor updates the accumulated physical displacement value in real time. When the processor detects that the effective material transport speed is zero and is in a shutdown state, the processor stops the increase of the accumulated physical displacement value and keeps the system clock counting continuously.

[0025] Furthermore, in step S2, the processor calculates the earliest and latest arrival times of the material at the downstream detection point based on the accumulated physical displacement value and the preset physical link design length, and constructs an effective kinematic time window, specifically including:

[0026] The processor reads the preset physical link design length and geometric path error rate, and calculates the Taylor dispersion length by combining the Taylor dispersion effect in fluid dynamics.

[0027] The processor calculates the difference between the numerical value and the geometric path error rate, and then calculates the product of the physical link design length and the difference to obtain the minimum physical transmission distance threshold.

[0028] The processor performs inverse kinematics retrieval in the time series of cumulative physical displacement values. The processor searches for the time index at which the cumulative physical displacement value first exceeds the minimum physical transmission distance threshold, and the processor locks the time index as the earliest arrival time of the material to the downstream detection point.

[0029] The processor calculates the sum of the numerical value and the geometric path error rate, the processor calculates the product of the physical link design length and the sum, the processor calculates the sum of the product and the Taylor dispersion length, and obtains the maximum physical transmission distance threshold.

[0030] The processor continues to perform inverse kinematic retrieval in the time series of the cumulative physical displacement value. The processor searches for the time index at which the cumulative physical displacement value first exceeds the maximum physical transmission distance threshold, and the processor locks the time index as the latest arrival time of the material to the downstream detection point.

[0031] The processor constructs a closed kinematically effective time window on the time axis using the earliest and latest arrival times.

[0032] Further, in step S3, the downstream sensor acquires the downstream response data sequence, and the processor uses the kinematic effective time window as the search space boundary to calculate the optimal time mapping path between the digitally encoded perturbation sequence and the downstream response data sequence using a physically constrained dynamic time warping algorithm. Specifically, this includes:

[0033] The processor performs robust spectral whitening operation based on physical noise floor locking, performs sliding window statistical operations on the downstream response data sequence, and calculates the local statistical mean and local statistical standard deviation of the downstream response data sequence within the sliding window.

[0034] The processor reads the sensor noise floor standard deviation preset in step S1, sets the signal-to-noise ratio lock-in coefficient, and calculates the product of the sensor noise floor standard deviation and the signal-to-noise ratio lock-in coefficient to obtain the noise floor lock-in threshold value.

[0035] The processor compares the values ​​of the local statistical standard deviation and the noise floor lockout threshold, and selects the larger value of the local statistical standard deviation and the noise floor lockout threshold as the denominator for anti-singularity regularization.

[0036] The processor calculates the difference between the downstream response data sequence and the local statistical mean. The processor divides the difference between the downstream response data sequence and the local statistical mean by the anti-singularity regularization denominator to generate a dimensionless normalized response sequence that eliminates gain fluctuations.

[0037] The processor performs normalization processing on the digitally encoded perturbation sequence to generate a normalized reference sequence;

[0038] The processor constructs a dynamic programming search grid. The processor strictly limits the search range of the dynamic programming search grid to the closed interval defined by the kinematic effective time window generated in step S2. The processor calculates the geometric Euclidean distance between the normalized reference sequence and the normalized response sequence at the grid nodes of the dynamic programming search grid. The processor uses the geometric Euclidean distance as the geometric cost component of the physical constraint dynamic time warping algorithm.

[0039] Furthermore, in step S3, the physical constraint dynamic time warping algorithm includes a velocity consistency penalty term. This penalty term is determined by the difference between the instantaneous transmission velocity sequence and the virtual flow rate ratio, where the virtual flow rate ratio is the local slope of the optimal time mapping path. Specifically, it includes:

[0040] The processor constructs a kinematic penalty field based on displacement conservation, obtains the rated design velocity of the system, and calculates the product of the local step size of the dynamic programming path on the reference sequence axis and the rated design velocity to obtain the virtual displacement at the source end.

[0041] The processor acquires the physically calibrated instantaneous transmission speed sequence output in step S2, and calculates the product of the local step size of the dynamic programming path on the response sequence axis and the instantaneous transmission speed sequence to obtain the physical displacement of the measurement end.

[0042] The processor calculates the difference between the virtual displacement at the source end and the physical displacement at the measurement end to obtain the displacement consistency residual;

[0043] The processor introduces an asymmetric relaxation potential well function to perform nonlinear mapping on the displacement consistency residual. When the physical displacement at the measurement end is less than the virtual displacement at the source end, the processor determines that the current state is a waveform compression behavior that violates the second law of thermodynamics, and the processor assigns a penalty value that increases the displacement consistency residual exponentially.

[0044] When the physical displacement at the measurement end is greater than the virtual displacement at the source end, the processor determines that the current state is a waveform broadening behavior that conforms to the Taylor dispersion effect. The processor assigns a penalty value that increases linearly with the displacement consistency residual and generates a velocity consistency penalty term.

[0045] The processor calculates the weighted sum of the geometric cost component and the speed consistency penalty term to obtain the local mixed potential energy. The processor then backtracks to generate the optimal time mapping path by minimizing the global cumulative mixed potential energy.

[0046] Further, in step S4, the processor uses the optimal time mapping path to map the data points in the downstream response data sequence back to the absolute physical input time. The processor obtains the material input mass flow rate corresponding to the absolute physical input time and performs a mass balance check, specifically including:

[0047] The processor performs an inverse projection operation based on the energy-weighted Dirac comb sieve. For the target feature time point in the downstream response data sequence, the processor retrieves all upstream reference time indices that are mapped to the target feature time point in the optimal time mapping path.

[0048] The processor calculates the local instantaneous power of the normalized reference sequence at each upstream reference time index, and uses the local instantaneous power of the signal as a weighting factor.

[0049] The processor calculates the product of all retrieved upstream reference time indices and the instantaneous power of the local signal corresponding to the upstream reference time index, and the processor calculates the sum of all products as the weighted time sum.

[0050] The processor calculates the sum of the instantaneous power of all retrieved local signals, and then uses the machine precision minimum value to regularize the sum to obtain a normalized weighted denominator.

[0051] The processor divides the weighted sum of times by the normalized weight denominator to calculate the absolute physical deployment time with high signal-to-noise ratio.

[0052] The processor uses the device calibration parameters to construct a disturbance flux conservation verification model. The processor calculates the product of the disturbance component in the physical tracer control command generated in step S1 and the preset actuator gain coefficient to obtain the upstream injection theoretical flux.

[0053] The processor calculates the downstream received inversion throughput based on the downstream response data sequence, preset sensor sensitivity, preset link transmission efficiency, and local Jacobian determinant.

[0054] The processor calculates the absolute value of the difference between the upstream injected theoretical flux and the downstream received inversion flux to obtain the flux imbalance residual.

[0055] The processor compares the throughput imbalance residual with a preset physical decision threshold. If the throughput imbalance residual is less than the preset physical decision threshold, the processor determines that the quality balance check has passed.

[0056] Furthermore, in step S4, the processor calculates the downstream received inversion throughput based on the downstream response data sequence, preset sensor sensitivity, preset link transmission efficiency, and local Jacobian determinant, specifically including:

[0057] The processor performs physical dimension restoration on the downstream response data sequence, divides the downstream response data sequence by the preset sensor sensitivity, and converts the electrical signal amplitude of the downstream response data sequence into a physical flow rate value.

[0058] The processor calculates the time derivative of the optimal time mapping path at the current moment to obtain the local Jacobian determinant, which is the spatiotemporal scaling factor during signal transmission.

[0059] The processor calculates the product of the physical flux rate value and the local Jacobian determinant, and performs a relativistic density compensation operation. The processor uses the relativistic density compensation operation to correct flux density changes caused by time compression or time stretching.

[0060] The processor introduces a link transmission efficiency parameter. The processor divides the data sequence after the relativistic density compensation operation by the link transmission efficiency parameter and performs a loss correction operation. The processor uses the loss correction operation to offset the energy attenuation caused by the pipe friction or chemical reaction.

[0061] The processor performs definite integral operations on the data sequence after density compensation and loss correction within the mapped time interval to calculate the downstream received inversion flux.

[0062] Furthermore, in step S4, the processor retrieves the responsible entity identifier from the production log based on the absolute physical input time, and generates a traceability and control data vector containing the responsible entity identifier, specifically including:

[0063] The processor performs a holographic spatiotemporal indexing operation, with the absolute physical input time as the time center point, and the processor constructs a time retrieval interval by combining the preset time tolerance range.

[0064] The processor obtains the cumulative physical displacement value calculated in step S2. The processor uses the cumulative physical displacement value corresponding to the absolute physical input time as the spatial center point. The processor combines the preset spatial tolerance range to construct a spatial retrieval interval.

[0065] The processor performs a dual-constraint retrieval operation in the production log of the manufacturing execution system, filtering out production records that simultaneously meet the time retrieval interval conditions and the spatial retrieval interval conditions; the processor extracts a unique batch number code or work group code from the filtered production records as the responsible entity identifier;

[0066] The processor encapsulates the responsible entity identifier, absolute physical input time, throughput imbalance residual, and traceability confidence into a traceability control data vector;

[0067] Among them, the source tracing confidence is a quantitative value calculated based on the flux imbalance residual and the global cumulative mixed potential energy. The source tracing confidence is used to numerically measure the credibility of the source tracing results in the physical conservation dimension and the waveform matching dimension.

[0068] The beneficial effects of this invention are as follows:

[0069] This invention overcomes the technical challenge of inaccurate traceability in non-steady-state production environments by actively implanting digitally encoded perturbation sequences as physical tracer fingerprints and combining them with kinematic modeling based on a Lagrange perspective. Unlike traditional statistical matching methods that rely solely on waveform geometric similarity, this invention constructs a physically-perceived temporal alignment mechanism that mandates that the time mapping path follows the laws of displacement conservation and entropy increase. Furthermore, it utilizes mass balance verification to physically validate the traceability results. This technological leap from signal correlation to physical causality effectively eliminates false alignment and quantization errors caused by downtime and speed changes, ensuring the absolute accuracy and immutability of identifying responsible entities and greatly enhancing the robustness and legal credibility of industrial data management systems. Attached Figure Description

[0070] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort:

[0071] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0072] The technical solutions of 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. Example

[0073] like Figure 1 As shown, this invention discloses a method for traceability and control of enterprise production data based on artificial intelligence, including the following steps:

[0074] Step S1: The controller obtains the basic process setting parameters and process safety constraint range of the current production process. The controller calculates the modulation depth threshold according to the process safety constraint range. The controller generates a digitally encoded disturbance sequence that meets the modulation depth threshold limit. The controller superimposes the digitally encoded disturbance sequence onto the basic process setting parameters, generates a physical tracer control command, and sends it to the production execution mechanism. The production execution mechanism drives material transfer according to the physical tracer control command.

[0075] This embodiment aims to solve the safety hazards caused by blindly injecting tracer signals in long-process continuous manufacturing, as well as the technical problems of signal physical filtering caused by the mismatch between signal frequency and actuator.

[0076] Specifically, in step S1, the controller obtains the basic process setting parameters and process safety constraint range of the current production process. The controller calculates the modulation depth threshold based on the process safety constraint range, which specifically includes:

[0077] The controller reads the pre-stored standard deviation of the sensor's background noise, and calculates the first physical distance between the basic process setting parameters and the upper limit of process safety in the process safety constraint range, as well as the second physical distance between the basic process setting parameters and the lower limit of process safety in the process safety constraint range.

[0078] In this embodiment, the basic process setting parameter is the set speed of the fluid transfer pump, and its physical unit is revolutions per minute; the process safety constraint range consists of the upper limit of process safety (e.g., 3,000 revolutions per minute) and the lower limit of process safety (e.g., 500 revolutions per minute).

[0079] The controller first reads the standard deviation of the sensor's noise floor. In this embodiment, the standard deviation of the sensor's noise floor is preferably set to five revolutions per minute. The selection of this value is based on the statistical analysis of one thousand consecutive sampling data of the speed sensor in a silent state, which represents the limit of physical observability under the current hardware conditions.

[0080] If the fluctuation caused by the injected signal is less than this value, it cannot be effectively detected physically. The controller calculates the absolute value of the difference between the basic process setting parameters and the upper limit of process safety as the first physical distance, and calculates the absolute value of the difference between the basic process setting parameters and the lower limit of process safety as the second physical distance, thereby quantifying the absolute margin of the current operating condition before physical collapse occurs.

[0081] The controller defines the smaller of the first physical distance and the second physical distance as the minimum safe distance, and compares the minimum safe distance with three times the standard deviation of the sensor's background noise.

[0082] In this embodiment, the controller performs a physical observability check based on the three sigma criterion. The controller selects the minimum of the first physical distance and the second physical distance as the minimum safe distance. The controller compares the minimum safe distance with three times the standard deviation of the sensor's noise floor (i.e., 15 revolutions per minute). Existing technologies usually ignore the influence of noise floor and directly inject signals, which results in the injected signal being completely submerged by noise when the process window is extremely narrow (e.g., the minimum safe distance is only 10 revolutions per minute). This not only makes it impossible to trace the source, but also increases the entropy of the control system. This embodiment establishes the minimum threshold for the physical signal to be observable by introducing a three-fold noise floor check.

[0083] When the minimum safe distance is less than three times the standard deviation of the sensor's noise floor, the controller sets the modulation depth threshold to zero. When the minimum safe distance is greater than or equal to three times the standard deviation of the sensor's noise floor, the controller calls the asymmetric safety barrier model to calculate the modulation depth threshold.

[0084] In this embodiment, when the minimum safe distance is less than 15 revolutions per minute, the controller triggers a hard physical cutoff logic, forcibly setting the modulation depth threshold to zero and stopping all signal injection, thereby avoiding the risk of ineffective operation under extreme conditions. When the minimum safe distance meets the observability requirements, the controller initiates the calculation of the asymmetric safety barrier model.

[0085] When the controller invokes the asymmetric security barrier model, it obtains the risk sensitivity length parameter, calculates the negative of the ratio of the minimum safe distance to the risk sensitivity length parameter, performs natural exponential operation on the negative of the ratio to obtain the exponential decay term, calculates the value minus the exponential decay term to obtain the security convergence factor, calculates the product of the minimum safe distance, the security convergence factor and the preset security margin coefficient, and assigns the result of the product to the modulation depth threshold to complete the calculation of the physical observability constraint of the modulation depth threshold.

[0086] In this embodiment, the logic for determining the risk sensitivity length parameter is as follows: The controller retrieves the enterprise's quality accident record database for the past three years, extracts the material impact range length corresponding to each quality anomaly event, fits all impact range length data to a log-normal distribution, calculates the upper quartile of the probability density function as the basic risk length, and multiplies the basic risk length by the risk sensitivity coefficient determined based on the unit value of the product (e.g., set to 1.2 for high-value products) to obtain the risk sensitivity length parameter. In this embodiment, the calculated risk sensitivity length parameter is set to 100 revolutions per minute. The physical meaning of this parameter is the physical span of the high-risk area. The controller calculates the negative of the ratio of the minimum safe distance to the risk sensitivity length parameter and performs natural exponential calculation to obtain the exponential decay term.

[0087] The mathematical model works as follows: when the basic process setting parameters are far from the boundary, the exponential decay term approaches zero, the safety convergence factor approaches one, and the system exhibits linear regulation; when the basic process setting parameters approach the boundary (i.e., entering the high-risk zone of 100 revolutions per minute), the exponential decay term increases rapidly, causing the safety convergence factor to decrease sharply and non-linearly. The controller multiplies the minimum safety distance, the safety convergence factor, and the safety margin coefficient (set to 0.3 in this embodiment, i.e., retaining a 70% safety margin) to obtain the modulation depth threshold.

[0088] Comparative experimental data shows that, compared with the fixed-ratio modulation method of the prior art, the asymmetric safety barrier model of this embodiment improves the convergence speed of the modulation depth threshold by 200% when the process parameters are close to the boundary, effectively avoiding downtime accidents caused by overshoot and significantly improving the inherent safety of the system.

[0089] In step S1, the controller generates a digitally coded perturbation sequence that meets the modulation depth threshold limit. The controller then superimposes the digitally coded perturbation sequence onto the basic process setting parameters to generate a physical tracer control command, specifically including:

[0090] The controller reads the step response time constant of the production actuator. The controller multiplies the step response time constant of the actuator by the bandwidth redundancy coefficient to obtain the minimum symbol holding time. The value of the bandwidth redundancy coefficient is not less than three.

[0091] In this embodiment, the production actuator is a high-inertia variable frequency motor, and its actuator step response time constant is 0.5 seconds. This parameter characterizes the physical time required for the motor speed to change from zero to 63% of the target value. Existing technologies often directly inject high-frequency pseudo-random sequences (e.g., with a frequency of 10 Hz), causing the actuator to be unable to respond due to mechanical inertia, and the signal is physically filtered out.

[0092] This embodiment introduces a bandwidth redundancy coefficient, which is set to five. The controller calculates that the minimum symbol holding time is two and a half seconds. The selection of this parameter is based on the physical extension of the Shannon sampling theorem, ensuring that the duration of each control symbol is sufficient for the actuator to complete a complete acceleration-stabilization-deceleration physical cycle, thus guaranteeing the integrity of the conversion of digital information into physical action.

[0093] The controller runs a discrete chaotic mapping algorithm to generate an original chaotic numerical sequence. The controller performs a sample-and-hold operation on the original chaotic numerical sequence based on the minimum symbol hold time. The controller performs symbolic binary processing on the numerical sequence after the sample-and-hold operation to generate a bandwidth-limited digital coded perturbation sequence.

[0094] In this embodiment, the controller uses Logistic mapping to generate an original chaotic numerical sequence with ergodicity. The controller samples and holds the original chaotic numerical sequence every 2.5 seconds (minimum symbol holding time) and maps the result to a binary logic of positive or negative one to generate a bandwidth-limited digital coded perturbation sequence. The spectral energy of this sequence is concentrated between zero and 0.2 Hz, which is completely within the effective response bandwidth of the actuator, thus solving the problem of high-frequency signal energy dissipation.

[0095] The controller calculates the product of the digitally encoded disturbance sequence and the modulation depth threshold to obtain the theoretical disturbance amplitude sequence. The controller performs an addition operation on the theoretical disturbance amplitude sequence and the basic process setting parameters to generate the original superimposed signal. The controller uses a dynamic smooth convolution kernel to perform time-domain convolution filtering on the original superimposed signal. The maximum value of the second derivative of the dynamic smooth convolution kernel is less than the maximum acceleration limit allowed by the production actuator.

[0096] In this embodiment, the maximum allowable acceleration limit of the production actuator is 50 revolutions per square second. If the original superimposed signal with a step change is directly sent, theoretically the motor is required to generate an infinite torque, which will lead to bearing wear. The controller uses a Hanning window as the dynamic smoothing convolution kernel. The maximum value of the second derivative of the convolution kernel is strictly designed to be 40 revolutions per square second. The controller performs convolution processing on the original superimposed signal, transforming the vertical signal edge into a continuous and differentiable S-shaped curve.

[0097] The controller eliminates the step abrupt edges in the original superimposed signal through time-domain convolution filtering, generating a continuous and differentiable smooth curve signal. The controller outputs the smooth curve signal as a physical tracer control command, and the production actuator drives the material transfer according to the physical tracer control command.

[0098] In this embodiment, the final generated physical tracer control command is a smooth speed change curve. After receiving the command, the production actuator drives the material conveying speed to generate controlled physical fluctuations. These fluctuations contain digital coding features that can be identified by downstream sensors, and are strictly limited within the process safety constraints and equipment mechanical strength limits.

[0099] In this embodiment, step S1 solves the technical problems of poor security, weak physical observability, and mismatch with the dynamic characteristics of the actuator in the tracer signal injection process by introducing an asymmetric safety barrier model and a bandwidth-limited dynamic shaping mechanism. Experimental data shows that this method increases the physical transmission success rate of the tracer signal from 40% to 98% in the prior art while ensuring that the yield rate is not affected. This provides a high signal-to-noise ratio physical source for the observation based on kinematic boundaries in the subsequent step S2, and is the physical basis for achieving accurate traceability throughout the entire process.

[0100] Step S2: The speed sensor collects the instantaneous transmission speed sequence on the material transmission link in real time. The processor performs time integration on the instantaneous transmission speed sequence to calculate the cumulative physical displacement value. Based on the cumulative physical displacement value and the preset physical link design length, the processor calculates the earliest and latest arrival times of the material to the downstream detection point and constructs an effective kinematic time window.

[0101] This embodiment aims to solve the technical problem that existing technologies cannot accurately predict the arrival time of materials under non-steady-state conditions such as production line speed changes, shutdowns, or buffering, thus causing the traceability algorithm to perform invalid searches at the wrong time. Existing technologies usually use a fixed time lag model (i.e., the physical link design length divided by the average speed equals the arrival time). This model assumes that the production speed is constant. However, in actual production, once equipment stops or speed fluctuates, the arrival time calculated by the fixed time model will deviate greatly from the actual physical arrival time, causing traceability to fail. This embodiment is based on the Lagrange perspective of material transport observation principle, and establishes a continuous medium kinematic model to define a rigid time boundary that conforms to the physical causality law.

[0102] Specifically, in step S2, the speed sensor collects the instantaneous transmission speed sequence on the material transport link in real time, and the processor performs time integration on the instantaneous transmission speed sequence to calculate the cumulative physical displacement value, which specifically includes:

[0103] The processor obtains the material rheological characteristic identifier of the current production batch from the manufacturing execution system. Based on the material rheological characteristic identifier, the processor queries the preset flow pattern distribution coefficient database and obtains the rheological correction factor that matches the current operating conditions.

[0104] In this embodiment, the sensor (such as a photoelectric encoder or electromagnetic flow meter) can usually only collect the mechanical rotation speed of the conveying equipment (such as a motor shaft or pulley), which is not equivalent to the actual movement speed of the material's center of mass.

[0105] The processor first identifies the material based on its rheological properties (e.g., high-viscosity lithium battery slurry) and matches the corresponding rheological correction factor in a pre-set flow pattern distribution coefficient database. Regarding the construction of this database, during the trial operation phase of the production line, the processor selects a typical test section. A Doppler ultrasonic velocimeter is installed at the central axis of the test section to measure the central velocity, while a high-precision Coriolis mass flow meter is installed at the pipe wall to measure the average mass flow rate. The processor adjusts the pumping pressure to ensure the material flow rate covers the laminar, transition, and turbulent regions. At each steady-state velocity point, the central velocity and average mass flow rate are recorded simultaneously, and their ratio is calculated. As the baseline correction factor, the processor uses Reynolds number and Prandtl number as independent variables and the baseline correction factor as the dependent variable, and performs polynomial regression analysis using the least squares method to generate the correction factor. In this embodiment, for low-viscosity fluids in turbulent conditions, the rheological correction factor is preferably set to 0.82. The value of this value is based on the analytical solution of the Navier-Stokes equations for the velocity distribution in a circular pipe cross section and the comparative calibration data of the on-site flow meter. The function of the rheological correction factor is to correct the excessively high central velocity caused by the laminar flow effect of the fluid, or the micro-slip phenomenon existing in the belt drive, and to calibrate the mechanical speed of the equipment to the net transport speed of the material.

[0106] Comparative data show that in the control group without the introduction of a rheological correction factor, the predicted material arrival time was delayed by an average of 15 percent due to the neglect of the laminar flow center velocity difference; while after the introduction of the rheological correction factor, the time prediction error was reduced to less than 2 percent.

[0107] The processor sets the zero-speed dead zone threshold of the sensor. The processor performs zero-speed dead zone logic judgment on the raw speed readings collected by the speed sensor. When the raw speed reading is lower than the sensor's zero-speed dead zone threshold, the processor determines that the effective material transport speed at the current moment is zero. When the raw speed reading is higher than the sensor's zero-speed dead zone threshold, the processor calculates the product of the raw speed reading and the rheological correction factor to obtain the physically calibrated effective material transport speed.

[0108] In this embodiment, in order to eliminate the slight reading drift caused by environmental vibration or circuit noise when the sensor is stationary, the processor introduces a sensor zero-speed dead zone threshold. The sensor zero-speed dead zone threshold means the maximum background noise limit of the sensor in the zero-input state. In this embodiment, the sensor zero-speed dead zone threshold is preferably set to 0.005 meters per second, and the value is based on the static noise statistics in the sensor technical specifications.

[0109] The processor performs logic gating operations: when the original speed reading is lower than the sensor's zero-speed dead zone threshold, the effective material transport speed is forcibly set to zero to prevent the integration of non-existent ghost mileage during long-term downtime; when the original speed reading is higher than the sensor's zero-speed dead zone threshold, the processor performs a multiplication operation, multiplying the original speed reading by the rheological correction factor, and outputting the effective material transport speed that conforms to physical facts.

[0110] Experimental verification shows that during a 12-hour intermittent production process, including a 4-hour downtime, without using the sensor zero-speed dead zone threshold, the false displacement generated by the accumulated background noise of the sensor can reach as high as 72 meters, which is enough to cause three batches to be misidentified in the traceability. However, by using the method of this embodiment, the false displacement is zero, ensuring the absolute accuracy of batch tracking.

[0111] The processor uses an anti-drift discrete Riemann sum algorithm to perform time-domain accumulation operation on the effective material transport speed. In each system sampling period, the processor calculates the product of the effective material transport speed and the system sampling period, and adds the product to the historical accumulated value. The processor updates the accumulated physical displacement value in real time. When the processor detects that the effective material transport speed is zero and is in a shutdown state, the processor stops the increase of the accumulated physical displacement value and keeps the system clock counting continuously.

[0112] In this embodiment, the processor uses the Discrete Riemann sum algorithm to convert the instantaneous velocity field into a cumulative displacement field. The cumulative physical displacement value represents the absolute distance that the material centroid actually moves in physical space from the moment the tracer signal is injected, in meters. The processor calculates and accumulates the infinitesimal displacement in each system sampling period (set to ten milliseconds in this embodiment).

[0113] In particular, this embodiment employs state-preserving logic: when the production line stops, causing the effective material transport speed to be zero, the cumulative physical displacement value remains unchanged, but the system clock continues to run. This logic mathematically replicates the process of material physically freezing in the pipeline, ensuring that no matter how many times the production line starts and stops, the correspondence between the cumulative physical displacement value and physical time always follows the principle of continuous medium mechanics and will not cause causal discontinuity.

[0114] In step S2, the processor calculates the earliest and latest arrival times of the material at the downstream detection point based on the cumulative physical displacement value and the preset physical link design length, and constructs an effective kinematic time window, specifically including:

[0115] The processor reads the preset physical link design length and geometric path error rate, and calculates the Taylor dispersion length by combining the Taylor dispersion effect in fluid dynamics.

[0116] In this embodiment, the processor first reads the physical link design length, that is, the geometric distance between the upstream and downstream sensors (500 meters in this embodiment), and introduces the geometric path error rate. The geometric path error rate characterizes the uncertainty of the path length caused by the thermal expansion and contraction of the pipeline or installation tolerances. In this embodiment, it is preferably set to one percent.

[0117] In addition, for fluid transport, the processor also needs to calculate the Taylor dispersion length, which characterizes the length of the ribbon-like tail formed by the tracer being stretched in the axial direction due to fluid viscosity. The processor obtains the Taylor dispersion length by calculating the square root of twice the product of the effective diffusion coefficient and the transport time. In the high-viscosity slurry transport scenario of this embodiment, the effective diffusion coefficient is set to 10 to the power of negative 9 square meters per second. The introduction of the Taylor dispersion length quantifies the entropy increase phenomenon in the material transport process, ensuring that the calculated time window can cover the slow particles trapped in the boundary layer.

[0118] The processor calculates the difference between the numerical value and the geometric path error rate, and then calculates the product of the physical link design length and the difference to obtain the minimum physical transmission distance threshold.

[0119] In this embodiment, the processor calculates the minimum physical transmission distance threshold through subtraction and multiplication operations. The minimum physical transmission distance threshold represents the shortest physical path that the material needs to take to reach the downstream, taking into account all negative errors.

[0120] The processor performs inverse kinematics retrieval in the time series of cumulative physical displacement values. The processor searches for the time index at which the cumulative physical displacement value first exceeds the minimum physical transmission distance threshold, and locks the time index as the earliest arrival time of the material to the downstream detection point.

[0121] In this embodiment, the processor searches the already generated sequence of cumulative physical displacement values ​​to find the moment when the value first breaks through the minimum physical transmission distance threshold. This moment is defined as the earliest arrival time, which means that even if the material is transported at the highest flow rate in the laminar flow center and has passed through the shortest geometric path, the tracer signal cannot arrive before this moment. This is the time left boundary of the causality.

[0122] The processor calculates the sum of the numerical value and the geometric path error rate, the processor calculates the product of the physical link design length and the sum, and the processor calculates the sum of the product and the Taylor dispersion length to obtain the maximum physical transmission distance threshold.

[0123] In this embodiment, the processor comprehensively considers the positive geometric error and the fluid dispersion effect to calculate the maximum physical transmission distance threshold. This calculation not only includes the upper limit of the path length, but also explicitly includes the Taylor dispersion length through addition, thereby reserving a reasonable physical space for the hysteresis particles.

[0124] Comparative experiments show that if only geometric errors are considered and Taylor dispersion length is ignored, in long-distance transmission scenarios (over 1,000 meters), about 18% of the lagging boundary layer signals will exceed the predicted latest arrival time, resulting in missed detections; after introducing Taylor dispersion length, the window coverage rate is improved to 99.9%.

[0125] The processor continues to perform inverse kinematic retrieval in the time series of cumulative physical displacement values. The processor searches for the time index at which the cumulative physical displacement value first exceeds the maximum physical transmission distance threshold, and locks the time index as the latest arrival time of the material to the downstream detection point.

[0126] In this embodiment, the processor continues to search the cumulative physical displacement value sequence to find the moment when the value first breaks through the maximum physical transmission distance threshold. This moment is defined as the latest arrival time, which is the right boundary of the causality in time and represents the limit time for all information-carrying particles to pass through the detection point.

[0127] The processor constructs a closed kinematically effective time window on the time axis using the earliest and latest arrival times.

[0128] In this embodiment, the processor closes the time interval between these two moments, forming a kinematically effective time window. The kinematically effective time window constitutes the causal horizon for subsequent data processing. Any time point outside the kinematically effective time window is physically considered an absolute vacuum, meaning that no real tracer signal can exist there.

[0129] Using the method in this embodiment, the processor successfully eliminated all false signals (artifacts) caused by sensor noise or environmental interference. In a production test involving frequent starts and stops, the search window determined by the traditional fixed delay method deviated from the actual signal arrival window by 240 seconds, resulting in the complete loss of the target. However, the kinematic effective time window constructed using this embodiment has an overlap of more than 98% with the actual signal arrival range, providing a physically hard-constrained search space for high-precision timing alignment in the subsequent step S3. This significantly reduces the computational complexity of the algorithm and improves the anti-interference capability of the source tracing.

[0130] Step S3: The downstream sensor collects the downstream response data sequence. The processor uses the kinematic effective time window as the search space boundary and uses the physical constraint dynamic time warping algorithm to calculate the optimal time mapping path between the digital coded perturbation sequence and the downstream response data sequence. The physical constraint dynamic time warping algorithm includes a velocity consistency penalty term, which is determined by the difference between the instantaneous transmission velocity sequence and the virtual flow velocity ratio. The virtual flow velocity ratio is the local slope of the optimal time mapping path.

[0131] This embodiment aims to address the technical problem in existing technologies during fluid transport or continuous manufacturing processes where fluctuations in transmission speed, equipment downtime, or media diffusion effects cause nonlinear distortions in the time axis between the upstream injected tracer signal and the downstream acquired response signal, leading to inaccurate matching. Traditional dynamic time warping algorithms only consider the similarity of waveform geometry, ignoring the kinematic constraints during physical transmission. This often results in ill-conditioned alignments that violate the laws of physical motion, such as forcibly stretching background noise during downtime to match the valid signal, or incorrectly matching waveforms from low-speed transmission segments with high-speed transmission segments. This embodiment achieves relativistic signal calibration in a non-inertial reference frame by introducing displacement conservation constraints and an asymmetric entropy increase relaxation mechanism, ensuring that the alignment path strictly follows the laws of continuous medium mechanics.

[0132] Specifically, in step S3, the downstream sensor acquires the downstream response data sequence, and the processor uses the kinematic effective time window as the search space boundary to calculate the optimal time mapping path between the digitally encoded perturbation sequence and the downstream response data sequence using a physically constrained dynamic time warping algorithm. This includes:

[0133] The processor performs robust spectral whitening based on physical noise floor locking, performs sliding window statistical operations on the downstream response data sequence, and calculates the local statistical mean and local statistical standard deviation of the downstream response data sequence within the sliding window.

[0134] In this embodiment, the length of the sliding window is set to fifty sampling points. This value is selected based on the Nyquist sampling analysis of the main frequency component of the digitally encoded perturbation sequence to ensure that the width of the sliding window is sufficient to cover at least two complete signal fluctuation cycles, thereby accurately extracting local statistical features. The processor calculates the local statistical mean within the sliding window to characterize the DC component and calculates the local statistical standard deviation to characterize the signal fluctuation energy.

[0135] The processor reads the sensor noise floor standard deviation preset in step S1, sets the signal-to-noise ratio lock-in coefficient, and calculates the product of the sensor noise floor standard deviation and the signal-to-noise ratio lock-in coefficient to obtain the noise floor lock-in threshold value.

[0136] In this embodiment, the standard deviation of the sensor's background noise is 0.01 MPa, and the signal-to-noise ratio lock-in coefficient is preferably set to 1.5. The value of the signal-to-noise ratio lock-in coefficient is based on the Ross criterion in the field of signal processing, that is, only when the signal strength exceeds 50% of the background noise does the signal have statistical significance for being distinguishable.

[0137] The processor compares the values ​​of the local statistical standard deviation and the noise floor lockout threshold. The processor selects the larger value of the local statistical standard deviation and the noise floor lockout threshold as the anti-singularity regularization denominator. This logic gating mechanism effectively solves the technical problem in the existing technology that when the production line is stopped or the signal is silent, the normalization calculation will have a division-by-zero error or the background noise will be amplified infinitely because the signal variance approaches zero.

[0138] The processor calculates the difference between the downstream response data sequence and the local statistical mean. The processor divides the difference between the downstream response data sequence and the local statistical mean by the anti-singularity regularization denominator to generate a dimensionless normalized response sequence that eliminates gain fluctuations. The processor performs normalization processing on the digitally encoded perturbation sequence to generate a normalized reference sequence.

[0139] Through the above processing, source signals and receiver signals with different physical dimensions and energy levels are uniformly mapped to a dimensionless statistical feature space, providing a unified mathematical benchmark for subsequent geometric morphology comparison.

[0140] The processor constructs a dynamic programming search grid, and strictly limits the search range of the dynamic programming search grid to the closed interval defined by the kinematic effective time window generated in step S2.

[0141] Comparative experimental data show that, compared with the traditional method of full-time domain search, this search space pruning strategy based on physical causality reduces the computational complexity of the algorithm by more than 95%, and fundamentally eliminates false matches (artifacts) outside the kinematic effective time window. The processor calculates the geometric Euclidean distance between the normalized reference sequence and the normalized response sequence at the grid nodes of the dynamic programming search grid, and the processor uses the geometric Euclidean distance as the geometric cost component of the physical constraint dynamic time warping algorithm.

[0142] In step S3, the physical constraint dynamic time warping algorithm includes a velocity consistency penalty term. This penalty term is determined by the difference between the instantaneous transmission velocity sequence and the virtual flow rate ratio, where the virtual flow rate ratio is the local slope of the optimal time mapping path. Specifically, it includes:

[0143] The processor constructs a kinematic penalty field based on displacement conservation, obtains the rated design velocity of the system, and calculates the product of the local step size of the dynamic programming path on the reference sequence axis and the rated design velocity to obtain the virtual displacement at the source end.

[0144] In this embodiment, the rated design speed is set to five meters per second, and the virtual displacement at the source end represents the physical distance that the material should theoretically move during the logical progression of the reference sequence.

[0145] The processor acquires the physically calibrated instantaneous transmission speed sequence output in step S2, and calculates the product of the local step size of the dynamic programming path on the response sequence axis and the instantaneous transmission speed sequence to obtain the physical displacement of the measurement end.

[0146] The physical displacement at the measuring end represents the actual distance the material flows during the actual physical transmission process. The processor calculates the difference between the virtual displacement at the source end and the physical displacement at the measuring end to obtain the displacement consistency residual. Under an ideal alignment path, the virtual displacement at the source end should be strictly equal to the physical displacement at the measuring end, that is, the displacement consistency residual is zero.

[0147] The processor introduces an asymmetric relaxation potential well function to perform nonlinear mapping on the displacement consistency residual. This function uses piecewise logic to adapt to physical laws: when the physical displacement at the measuring end is less than the virtual displacement at the source end, the processor determines that the current state is waveform compression behavior that violates the second law of thermodynamics. At this time, the function adopts an exponential penalty form, and the penalty value is equal to the power of Euler's constant. The exponent is the product of the absolute value of the displacement consistency residual and the first sensitivity coefficient (set to ten in this embodiment), thereby giving the displacement consistency residual an exponentially increasing penalty value, forming a hard constraint; when the physical displacement at the measuring end is greater than the virtual displacement at the source end, the processor determines that the current state is waveform broadening behavior that conforms to the Taylor dispersion effect. At this time, the function adopts a linear penalty form, and the penalty value is equal to the product of the displacement consistency residual and the second sensitivity coefficient (set to 0.1 in this embodiment), giving the displacement consistency residual a linearly increasing penalty value, generating a velocity consistency penalty term, thereby forming a soft compatibility interval.

[0148] Physical analysis shows that when the physical displacement at the measuring end is less than the virtual displacement at the source end, it means that the algorithm is trying to compress a long original signal into a short physical displacement. This physically implies that the speed of signal propagation exceeds the speed of material transport, violating causality. Therefore, the processor forms a hard constraint through exponential penalties to forcibly prohibit the generation of such paths. This mechanism is adapted to the shutdown condition: when the instantaneous transmission speed sequence is zero, the physical displacement at the measuring end is zero, forcing the virtual displacement at the source end to be zero (i.e., the logical step size stops advancing), thus achieving synchronous freezing with physical shutdown at the algorithm level.

[0149] When the physical displacement at the measurement end is greater than the virtual displacement at the source end, the processor determines that the current state conforms to the waveform broadening behavior of the Taylor dispersion effect. The processor assigns a penalty value that increases linearly with the displacement consistency residual and generates a velocity consistency penalty term.

[0150] Fluid dynamics theory shows that due to the Taylor dispersion effect caused by fluid viscosity, the signal will inevitably undergo a certain degree of waveform broadening (entropy increase) during transmission. Therefore, the processor only applies a linearly increasing soft penalty, allowing time axis stretching within the range that conforms to the physical diffusion law.

[0151] The processor calculates the weighted sum of the geometric cost component and the speed consistency penalty term to obtain the local mixed potential energy. The processor then backtracks to generate the optimal time mapping path by minimizing the global cumulative mixed potential energy.

[0152] Comparative tests show that when processing complex operating data containing frequent speed changes (speed fluctuations exceeding 50%) and intermittent shutdowns, the alignment accuracy of the physical constraint dynamic time warping algorithm in this embodiment is improved by two orders of magnitude compared to the traditional algorithm that does not introduce a speed consistency penalty term.

[0153] Specifically, traditional algorithms have an average time alignment error of 500 milliseconds during the speed change phase and generate severe signal aliasing during the stop phase. In contrast, the method in this embodiment controls the time alignment error to within 10 milliseconds and maintains strict logical stillness during the stop phase, without generating any false matches. This technical effect demonstrates the decisive role of introducing physical constraints in improving the reliability of tracing in non-steady-state environments.

[0154] In this embodiment, step S3 constructs a physical sensing time alignment mechanism and applies strict continuous medium mechanical constraints to the dynamic programming process using instantaneous transmission speed sequences. This solves the signal distortion calibration problem in a non-inertial reference frame. The optimal time mapping path output by this step provides a unique and reliable mathematical mapping benchmark for the subsequent step S4 to accurately restore the chaotic downstream data to the absolute physical time axis. This is a key link in realizing the leap from data correlation to physical causality.

[0155] Step S4: The processor uses the optimal time mapping path to map the data points in the downstream response data sequence back to the absolute physical input time. The processor obtains the material input quality flow rate corresponding to the absolute physical input time and performs a quality balance check. After passing the quality balance check, the processor retrieves the responsible entity identifier in the production log based on the absolute physical input time and generates a traceability control data vector containing the responsible entity identifier.

[0156] This embodiment aims to solve the technical problem that the existing technology, after completing time alignment, relies solely on waveform geometric similarity for liability determination, which is prone to false alarms or misreports due to accidental mathematical correlations. In complex industrial production environments, different batches may generate wave signals with similar shapes. If the determination is based solely on the distance calculated by the dynamic time warping algorithm, it is very easy to mistakenly associate two completely unrelated production events.

[0157] This embodiment is based on Noether's conservation law and continuity equation in physics. It introduces the conservation of perturbation flux as a hard physical criterion to verify whether the energy injected upstream and the energy received downstream are still conserved after undergoing spatiotemporal transformation, thereby eliminating false matches and realizing the leap from signal correlation to physical causality.

[0158] Specifically, in step S4, the processor uses the optimal time mapping path to map data points in the downstream response data sequence back to the absolute physical input time. The processor obtains the material input mass flow rate corresponding to the absolute physical input time and performs a mass balance check, which specifically includes:

[0159] The processor performs a reverse projection operation based on the energy-weighted Dirac comb sieve. For the target feature time point in the downstream response data sequence, the processor retrieves all upstream reference time indices that are mapped to the target feature time point in the optimal time mapping path.

[0160] In this embodiment, the optimal time mapping path is typically represented as a discrete set of grid points. Because the dynamic time warping algorithm allows for nonlinear scaling of the time axis, one downstream time point may correspond to multiple upstream time points (e.g., in the signal compression segment), or multiple downstream time points may correspond to one upstream time point (e.g., in the shutdown stretching segment). Direct lookup table inversion can lead to quantization errors in time positioning.

[0161] The processor first identifies a significant peak in the downstream response data sequence whose amplitude exceeds three times the background noise as the target feature time point, and then retrieves all upstream reference time index sets pointing to the target feature time point from the path set.

[0162] The processor calculates the local instantaneous power of the normalized reference sequence at each upstream reference time index, and uses the local instantaneous power as a weighting factor.

[0163] In this embodiment, the processor reads the normalized reference sequence, which serves as the source template, and calculates its squared value at each index position to obtain the instantaneous power of the local signal. The significance of this step is to use the energy distribution characteristics of the signal to evaluate the confidence of time reversal. According to the signal-to-noise ratio theory, regions with higher signal energy (such as pulse transition edges) have stronger anti-noise interference capabilities and more accurate time positioning; while low-energy regions (such as flat segments) are easily affected by random noise and produce phase drift. Therefore, this embodiment selects the instantaneous power of the local signal as a weighting factor to give higher-energy feature points greater weight.

[0164] The processor calculates the product of all retrieved upstream reference time indices and the corresponding instantaneous power of the local signal. The processor calculates the sum of all products as the weighted time sum. The processor calculates the sum of all retrieved instantaneous power of the local signal. The processor introduces the minimum value of machine precision to perform regularization on the sum to obtain the normalized weight denominator.

[0165] In this embodiment, the processor performs a weighted summation operation. To prevent division by zero errors or numerical instability caused by the instantaneous power of the local signal approaching zero at signal zero-crossing points or in silent regions, the processor introduces a minimum machine precision value (set to 10 to the power of negative 9 in this embodiment) to regularize the accumulated sum. This processing ensures that the normalized weight denominator is a non-zero positive number under any operating condition, guaranteeing the robustness of the algorithm.

[0166] The processor divides the weighted sum of times by the normalized weight denominator to calculate the absolute physical deployment time with a high signal-to-noise ratio.

[0167] In this embodiment, the processor calculates the absolute physical input time using the energy centroid method described above. Comparative experimental data shows that when processing noisy signals with a signal-to-noise ratio of 10 dB, the average time positioning error is 50 milliseconds when using the traditional direct lookup table method for time inversion. However, when using the energy-weighted Dirac comb sieve reverse projection operation of this embodiment, the average time positioning error is reduced to 2 milliseconds, significantly improving the accuracy of time inversion and eliminating the quantization jitter caused by the discrete grid.

[0168] The processor uses the device calibration parameters to construct a disturbance flux conservation verification model. The processor calculates the product of the disturbance component in the physical tracer control command generated in step S1 and the preset actuator gain coefficient to obtain the upstream injection theoretical flux.

[0169] In this embodiment, the processor begins to perform a verification based on the physical conservation law. The processor extracts the disturbance voltage amplitude (unit: volts) from the physical tracer control instruction and multiplies it by a preset actuator gain coefficient.

[0170] In this embodiment, for a screw pump driven by a variable frequency drive, the actuator gain coefficient is calibrated to 0.5 kg / s per volt. This parameter is derived from the flow-voltage characteristic curve of the equipment at the time of manufacture. The product of the two is the theoretical upstream injection flux, which characterizes the tracer mass flow rate of the injection system in theory.

[0171] The processor calculates the downstream received inversion throughput based on the downstream response data sequence, preset sensor sensitivity, preset link transmission efficiency, and local Jacobian determinant.

[0172] In this embodiment, the processor calculates the actual physical throughput received downstream by inversion.

[0173] The processor calculates the absolute value of the difference between the upstream injected theoretical flux and the downstream received inversion flux to obtain the flux imbalance residual. The processor compares the flux imbalance residual with a preset physical decision threshold. If the flux imbalance residual is less than the preset physical decision threshold, the processor determines that the quality balance check has passed.

[0174] In this embodiment, the processor calculates the absolute value of the difference between the two to obtain the flux imbalance residual. This parameter quantifies the degree to which the traceability result violates the law of conservation of mass. The preset physical decision threshold is set to five percent of the theoretical flux injected upstream. The selection of this threshold is based on the industrial process metrology level standard (Class 0.5) and the statistical value of unmeasurable disturbances in the fluid transport process. If the flux imbalance residual is less than this threshold, it indicates that the waveform matching is not only similar in shape, but also has a solid material transport basis, thus the verification is judged to be passed; otherwise, it is regarded as a false match and the responsibility is rejected.

[0175] In step S4, the processor calculates the downstream received inversion throughput based on the downstream response data sequence, preset sensor sensitivity, preset link transmission efficiency, and local Jacobian determinant, specifically including:

[0176] The processor performs physical dimension restoration on the downstream response data sequence, divides the downstream response data sequence by the preset sensor sensitivity, and converts the electrical signal amplitude of the downstream response data sequence into a physical flow rate value.

[0177] In this embodiment, the downstream sensor (such as a piezoelectric pressure sensor) outputs a voltage signal, which cannot be directly compared with the mass flow rate. The processor divides it by a preset sensor sensitivity, which in this embodiment is calibrated to 10 volts per kilogram per second. This step completes the conversion from electrical dimensions to mechanical dimensions.

[0178] The processor calculates the time derivative of the optimal time mapping path at the current moment, and obtains the local Jacobian determinant, which is the spatiotemporal scaling factor during signal transmission.

[0179] In this embodiment, the processor performs differential estimation on the discrete optimal time mapping path to obtain the local Jacobian determinant, which represents the scaling factor of the time coordinate transformation. For example, a value of 0.5 indicates that the time is compressed by half, corresponding to a faster flow rate; a value of 2 indicates that the time is stretched by half, corresponding to a slower flow rate.

[0180] The processor calculates the product of the physical flux rate and the local Jacobian determinant, and performs a relativistic density compensation operation. The processor uses the relativistic density compensation operation to correct flux density changes caused by time compression or time stretching.

[0181] In this embodiment, this is a key step in verifying the conservation of matter. According to the principle of continuum mechanics, when the material transport speed increases and the time axis is compressed, the mass flux density per unit time will inevitably increase. If only the original waveform is integrated, the calculated total mass will be too small. Therefore, the processor must multiply the physical flux value by the local Jacobian determinant to perform relativistic density compensation.

[0182] Comparative experiments show that under unsteady conditions where the flow velocity fluctuation exceeds 50%, if this step is ignored, the quality integration error of the high-speed transmission segment will exceed 40%, causing the system to erroneously intercept the correct tracing results; however, after introducing this compensation, the integration error is controlled within 2%.

[0183] The processor introduces a link transmission efficiency parameter. The processor divides the data sequence after the relativistic density compensation operation by the link transmission efficiency parameter and performs a loss correction operation. The processor uses the loss correction operation to offset the energy attenuation caused by pipeline friction or chemical reactions.

[0184] When performing loss correction operations, considering friction loss or natural loss of volatile materials during long-distance pipeline transmission, the processor introduces a link transmission efficiency parameter. This parameter is calibrated as follows: the processor uses a tracer pulse response test method, controlling the upstream feeder to inject a known mass of standard tracer salt solution within a very short time to form a Dirac pulse input. Downstream sensors continuously monitor conductivity changes until the signal recovers to the baseline. The processor performs time integration on the conductivity change curve collected downstream to calculate the total mass of tracer flowing through, and calculates the ratio of the total mass recovered downstream to the total mass injected upstream. In this embodiment, based on the above measured calibration, this parameter is set to 0.95, representing 5% physical loss. The processor compensates for the data through division operations to restore the theoretical received amount under lossless conditions.

[0185] The processor performs definite integral operations on the data sequence after density compensation and loss correction within the mapped time interval to calculate the downstream received inversion flux.

[0186] In this embodiment, the processor performs a definite integral on the processed data sequence within the mapped time interval and calculates the area under the curve. This area is the downstream received inversion flux, which represents the total mass of material observed downstream and reverted upstream.

[0187] In step S4, the processor retrieves the responsible entity identifier from the production log based on the absolute physical input time, and generates a traceability and control data vector containing the responsible entity identifier, specifically including:

[0188] The processor performs a holographic spatiotemporal indexing operation, using the absolute physical input time as the time center point, and constructs a time retrieval interval by combining the preset time tolerance range.

[0189] In this embodiment, the processor first constructs the retrieval conditions in the time dimension. Taking the calculated absolute physical input time as the center, a preset time tolerance range is introduced. In this embodiment, the range is set to plus or minus two seconds, and the value is based on the time span corresponding to the maximum tail length calculated based on the Taylor dispersion effect in step S2.

[0190] The processor obtains the cumulative physical displacement value calculated in step S2. The processor uses the cumulative physical displacement value corresponding to the absolute physical input time as the spatial center point. The processor constructs a spatial retrieval interval by combining the preset spatial tolerance range.

[0191] In this embodiment, in order to prevent erroneous indexes caused by system clock deviations (such as NTP synchronization failure), the processor introduces spatial dimension verification. The processor reads the cumulative physical displacement value obtained by integration in step S2 and constructs a spatial retrieval interval by combining it with a preset spatial tolerance range (set to ±5 meters in this embodiment, determined based on the cumulative error statistics of the encoder during long-term operation).

[0192] The processor performs a dual-constraint retrieval operation in the production log of the manufacturing execution system. The processor filters out production records that simultaneously meet the time retrieval interval conditions and the spatial retrieval interval conditions. The processor extracts a unique batch number code or work group code from the filtered production records as the responsible entity identifier.

[0193] In this embodiment, the processor performs a spatiotemporal intersection search in the database. Only when a production record falls within the above interval in both time and space is it locked. This completely eliminates erroneous attribution caused by a single time index deviation.

[0194] For example, in a test, a three-second clock deviation occurred due to network latency. Searching solely by time would incorrectly point to the next batch. However, by introducing a spatial search interval (odometer data is stored locally on the controller and is unaffected by network latency), the processor successfully corrected the deviation and locked onto the correct production record. The processor then extracted the batch number code from this record as the identifier of the responsible entity.

[0195] The processor encapsulates the responsible entity identifier, absolute physical input time, flux imbalance residual, and traceability confidence into a traceability control data vector. The traceability confidence is a quantitative value calculated based on the flux imbalance residual and the global cumulative mixed potential energy. The traceability confidence is used to numerically measure the credibility of the traceability results in the physical conservation dimension and the waveform matching dimension.

[0196] In this embodiment, the processor generates the final traceability and control data vector. The traceability confidence is obtained by weighting the flux imbalance residual (physical dimension consistency) and the global cumulative mixed potential (mathematical dimension similarity) using a negative exponential function, and outputs a dimensionless value between zero and one. This vector, as tamper-proof electronic evidence, is transmitted to the enterprise control platform for quality traceability, responsibility identification, and process optimization.

[0197] In this embodiment, step S4 establishes a rigid coupling relationship between physical flux and signal waveform, and uses Noether's conservation law to perform a physical-level final verification of the output of the mathematical algorithm. This method not only achieves accurate time inversion, but more importantly, it provides a counterfeit-proof verification mechanism based on physical iron laws, ensuring the absolute credibility of the traceability results in terms of laws and industry standards, and completing the closed loop of full-process data traceability control.

[0198] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for traceability and control of enterprise production data based on artificial intelligence, characterized in that, Includes the following steps: Step S1: The controller obtains the basic process setting parameters and process safety constraint range of the current production process. The controller calculates the modulation depth threshold according to the process safety constraint range. The controller generates a digitally encoded disturbance sequence that meets the modulation depth threshold limit. The controller superimposes the digitally encoded disturbance sequence onto the basic process setting parameters, generates a physical tracer control command, and sends it to the production execution mechanism. The production execution mechanism drives material transfer according to the physical tracer control command. Step S2: The speed sensor collects the instantaneous transmission speed sequence on the material transmission link in real time. The processor performs time integration on the instantaneous transmission speed sequence to calculate the cumulative physical displacement value. Based on the cumulative physical displacement value and the preset physical link design length, the processor calculates the earliest and latest arrival times of the material to the downstream detection point and constructs an effective kinematic time window. Step S3: The downstream sensor collects the downstream response data sequence. The processor uses the kinematic effective time window as the search space boundary and uses the physical constraint dynamic time warping algorithm to calculate the optimal time mapping path between the digitally encoded perturbation sequence and the downstream response data sequence. The physical constraint dynamic time warping algorithm includes a velocity consistency penalty term, which is determined by the difference between the instantaneous transmission velocity sequence and the virtual flow velocity ratio. The virtual flow velocity ratio is the local slope of the optimal time mapping path. In step S3, the downstream sensor acquires the downstream response data sequence. The processor uses the kinematic effective time window as the search space boundary and employs a physically constrained dynamic time warping algorithm to calculate the optimal time mapping path between the digitally encoded perturbation sequence and the downstream response data sequence. Specifically, this includes: The processor performs robust spectral whitening operation based on physical noise floor locking, performs sliding window statistical operations on the downstream response data sequence, and calculates the local statistical mean and local statistical standard deviation of the downstream response data sequence within the sliding window. The processor reads the sensor noise floor standard deviation preset in step S1, sets the signal-to-noise ratio lock-in coefficient, and calculates the product of the sensor noise floor standard deviation and the signal-to-noise ratio lock-in coefficient to obtain the noise floor lock-in threshold value. The processor compares the values ​​of the local statistical standard deviation and the noise floor lockout threshold, and selects the larger value of the local statistical standard deviation and the noise floor lockout threshold as the denominator for anti-singularity regularization. The processor calculates the difference between the downstream response data sequence and the local statistical mean. The processor divides the difference between the downstream response data sequence and the local statistical mean by the anti-singularity regularization denominator to generate a dimensionless normalized response sequence that eliminates gain fluctuations. The processor performs normalization processing on the digitally encoded perturbation sequence to generate a normalized reference sequence; The processor constructs a dynamic programming search grid. The processor strictly limits the search range of the dynamic programming search grid to the closed interval defined by the kinematic effective time window generated in step S2. The processor calculates the geometric Euclidean distance between the normalized reference sequence and the normalized response sequence at the grid nodes of the dynamic programming search grid. The processor uses the geometric Euclidean distance as the geometric cost component of the physical constraint dynamic time warping algorithm. Step S4: The processor uses the optimal time mapping path to map the data points in the downstream response data sequence back to the absolute physical input time. The processor obtains the material input quality flow rate corresponding to the absolute physical input time and performs a quality balance check. After passing the quality balance check, the processor retrieves the responsible entity identifier in the production log based on the absolute physical input time and generates a traceability control data vector containing the responsible entity identifier.

2. The method for traceability and control of enterprise production data based on artificial intelligence according to claim 1, characterized in that, In step S1, the controller calculates the modulation depth threshold based on the process safety constraints, specifically including: The controller reads the pre-stored standard deviation of the sensor's background noise, and calculates the first physical distance between the basic process setting parameters and the upper limit of process safety in the process safety constraint range, and the second physical distance between the basic process setting parameters and the lower limit of process safety in the process safety constraint range. The controller defines the smaller of the first physical distance and the second physical distance as the minimum safe distance, and compares the minimum safe distance with three times the standard deviation of the sensor's background noise. When the minimum safe distance is less than three times the standard deviation of the sensor's noise floor, the controller sets the modulation depth threshold value directly to zero. When the minimum safe distance is greater than or equal to three times the standard deviation of the sensor's noise floor, the controller calls the asymmetric safety barrier model to calculate the modulation depth threshold. When the controller invokes the asymmetric security barrier model, it obtains the risk sensitivity length parameter, calculates the negative of the ratio of the minimum safe distance to the risk sensitivity length parameter, performs natural exponential operation on the negative of the ratio to obtain the exponential decay term, calculates the value minus the exponential decay term to obtain the security convergence factor, calculates the product of the minimum safe distance, the security convergence factor and the preset security margin coefficient, and assigns the result of the product to the modulation depth threshold to complete the calculation of the physical observability constraint of the modulation depth threshold.

3. The method for traceability and control of enterprise production data based on artificial intelligence according to claim 2, characterized in that, In step S1, the controller generates a digitally coded perturbation sequence that meets the modulation depth threshold limit. The controller then superimposes the digitally coded perturbation sequence onto the basic process setting parameters to generate a physical tracer control command, specifically including: The controller reads the step response time constant of the production actuator, and multiplies the step response time constant by the bandwidth redundancy coefficient to obtain the minimum symbol holding time. The value of the bandwidth redundancy coefficient is not less than three. The controller runs a discrete chaotic mapping algorithm to generate an original chaotic numerical sequence. The controller performs a sample-and-hold operation on the original chaotic numerical sequence based on the minimum symbol holding time. The controller performs symbolic binary processing on the numerical sequence after the sample-and-hold operation to generate a bandwidth-limited digital coded perturbation sequence. The controller calculates the product of the digitally encoded disturbance sequence and the modulation depth threshold to obtain the theoretical disturbance amplitude sequence. The controller performs an addition operation on the theoretical disturbance amplitude sequence and the basic process setting parameters to generate the original superimposed signal. The controller uses a dynamic smooth convolution kernel to perform time-domain convolution filtering on the original superimposed signal. The maximum value of the second derivative of the dynamic smooth convolution kernel is less than the maximum allowable acceleration limit of the production actuator. The controller eliminates the step abruption edges in the original superimposed signal through time-domain convolution filtering, generating a continuous and differentiable smooth curve signal. The controller outputs the smooth curve signal as a physical tracer control command.

4. The method for traceability and control of enterprise production data based on artificial intelligence according to claim 1, characterized in that, In step S2, the speed sensor collects the instantaneous transmission speed sequence on the material transport link in real time, and the processor performs time integration on the instantaneous transmission speed sequence to calculate the cumulative physical displacement value, specifically including: The processor obtains the material rheological characteristic identifier of the current production batch from the manufacturing execution system. Based on the material rheological characteristic identifier, the processor queries the preset flow pattern distribution coefficient database and obtains the rheological correction factor that matches the current operating conditions. The processor sets the zero-speed dead zone threshold of the sensor. The processor performs zero-speed dead zone logic judgment on the raw speed readings collected by the speed sensor. When the raw speed reading is lower than the zero-speed dead zone threshold of the sensor, the processor determines that the effective material transport speed at the current moment is zero. When the raw speed reading is higher than the zero-speed dead zone threshold of the sensor, the processor calculates the product of the raw speed reading and the rheological correction factor to obtain the physically calibrated effective material transport speed. The processor uses an anti-drift discrete Riemann sum algorithm to perform time-domain accumulation operation on the effective material transport speed. In each system sampling period, the processor calculates the product of the effective material transport speed and the system sampling period, and adds the product to the historical accumulated value. The processor updates the accumulated physical displacement value in real time. When the processor detects that the effective material transport speed is zero and is in a shutdown state, the processor stops the increase of the accumulated physical displacement value and keeps the system clock counting continuously.

5. The method for traceability and control of enterprise production data based on artificial intelligence according to claim 4, characterized in that, In step S2, the processor calculates the earliest and latest arrival times of the material at the downstream detection point based on the cumulative physical displacement value and the preset physical link design length, and constructs an effective kinematic time window, specifically including: The processor reads the preset physical link design length and geometric path error rate, and calculates the Taylor dispersion length by combining the Taylor dispersion effect in fluid dynamics. The processor calculates the difference between the numerical value and the geometric path error rate, and then calculates the product of the physical link design length and the difference to obtain the minimum physical transmission distance threshold. The processor performs inverse kinematics retrieval in the time series of cumulative physical displacement values. The processor searches for the time index at which the cumulative physical displacement value first exceeds the minimum physical transmission distance threshold, and the processor locks the time index as the earliest arrival time of the material to the downstream detection point. The processor calculates the sum of the numerical value and the geometric path error rate, the processor calculates the product of the physical link design length and the sum, the processor calculates the sum of the product and the Taylor dispersion length, and obtains the maximum physical transmission distance threshold. The processor continues to perform inverse kinematic retrieval in the time series of the cumulative physical displacement value. The processor searches for the time index at which the cumulative physical displacement value first exceeds the maximum physical transmission distance threshold, and the processor locks the time index as the latest arrival time of the material to the downstream detection point. The processor constructs a closed kinematically effective time window on the time axis using the earliest and latest arrival times.

6. The method for traceability and control of enterprise production data based on artificial intelligence according to claim 5, characterized in that, In step S3, the physical constraint dynamic time warping algorithm includes a velocity consistency penalty term. This penalty term is determined by the difference between the instantaneous transmission velocity sequence and the virtual flow rate ratio, where the virtual flow rate ratio is the local slope of the optimal time mapping path. Specifically, it includes: The processor constructs a kinematic penalty field based on displacement conservation, obtains the rated design velocity of the system, and calculates the product of the local step size of the dynamic programming path on the reference sequence axis and the rated design velocity to obtain the virtual displacement at the source end. The processor acquires the physically calibrated instantaneous transmission speed sequence output in step S2, and calculates the product of the local step size of the dynamic programming path on the response sequence axis and the instantaneous transmission speed sequence to obtain the physical displacement of the measurement end. The processor calculates the difference between the virtual displacement at the source end and the physical displacement at the measurement end to obtain the displacement consistency residual; The processor introduces an asymmetric relaxation potential well function to perform nonlinear mapping on the displacement consistency residual. When the physical displacement at the measurement end is less than the virtual displacement at the source end, the processor determines that the current state is a waveform compression behavior that violates the second law of thermodynamics, and the processor assigns a penalty value that increases the displacement consistency residual exponentially. When the physical displacement at the measurement end is greater than the virtual displacement at the source end, the processor determines that the current state is a waveform broadening behavior that conforms to the Taylor dispersion effect. The processor assigns a penalty value that increases linearly with the displacement consistency residual and generates a velocity consistency penalty term. The processor calculates the weighted sum of the geometric cost component and the speed consistency penalty term to obtain the local mixed potential energy. The processor then backtracks to generate the optimal time mapping path by minimizing the global cumulative mixed potential energy.

7. The method for traceability and control of enterprise production data based on artificial intelligence according to claim 1, characterized in that, In step S4, the processor uses the optimal time mapping path to map data points in the downstream response data sequence back to the absolute physical input time. The processor obtains the material input mass flow rate corresponding to the absolute physical input time and performs a mass balance check, specifically including: The processor performs an inverse projection operation based on the energy-weighted Dirac comb sieve. For the target feature time point in the downstream response data sequence, the processor retrieves all upstream reference time indices that are mapped to the target feature time point in the optimal time mapping path. The processor calculates the local instantaneous power of the normalized reference sequence at each upstream reference time index, and uses the local instantaneous power of the signal as a weighting factor. The processor calculates the product of all retrieved upstream reference time indices and the instantaneous power of the local signal corresponding to the upstream reference time index, and the processor calculates the sum of all products as the weighted time sum. The processor calculates the sum of the instantaneous power of all retrieved local signals, and then uses the machine precision minimum value to regularize the sum to obtain a normalized weighted denominator. The processor divides the weighted sum of times by the normalized weight denominator to calculate the absolute physical deployment time with high signal-to-noise ratio. The processor uses the device calibration parameters to construct a disturbance flux conservation verification model. The processor calculates the product of the disturbance component in the physical tracer control command generated in step S1 and the preset actuator gain coefficient to obtain the upstream injection theoretical flux. The processor calculates the downstream received inversion throughput based on the downstream response data sequence, preset sensor sensitivity, preset link transmission efficiency, and local Jacobian determinant. The processor calculates the absolute value of the difference between the upstream injected theoretical flux and the downstream received inversion flux to obtain the flux imbalance residual. The processor compares the throughput imbalance residual with a preset physical decision threshold. If the throughput imbalance residual is less than the preset physical decision threshold, the processor determines that the quality balance check has passed.

8. The method for traceability and control of enterprise production data based on artificial intelligence according to claim 7, characterized in that, In step S4, the processor calculates the downstream received inversion throughput based on the downstream response data sequence, preset sensor sensitivity, preset link transmission efficiency, and local Jacobian determinant, specifically including: The processor performs physical dimension restoration on the downstream response data sequence, divides the downstream response data sequence by the preset sensor sensitivity, and converts the electrical signal amplitude of the downstream response data sequence into a physical flow rate value. The processor calculates the time derivative of the optimal time mapping path at the current moment to obtain the local Jacobian determinant, which is the spatiotemporal scaling factor during signal transmission. The processor calculates the product of the physical flux rate value and the local Jacobian determinant, and performs a relativistic density compensation operation. The processor uses the relativistic density compensation operation to correct flux density changes caused by time compression or time stretching. The processor introduces a link transmission efficiency parameter. The processor divides the data sequence after the relativistic density compensation operation by the link transmission efficiency parameter and performs a loss correction operation. The processor uses the loss correction operation to offset the energy attenuation caused by the pipe friction or chemical reaction. The processor performs definite integral operations on the data sequence after density compensation and loss correction within the mapped time interval to calculate the downstream received inversion flux.

9. The method for traceability and control of enterprise production data based on artificial intelligence according to claim 8, characterized in that, In step S4, the processor retrieves the responsible entity identifier from the production log based on the absolute physical input time, and generates a traceability and control data vector containing the responsible entity identifier, specifically including: The processor performs a holographic spatiotemporal indexing operation, with the absolute physical input time as the time center point, and the processor constructs a time retrieval interval by combining the preset time tolerance range. The processor obtains the cumulative physical displacement value calculated in step S2. The processor uses the cumulative physical displacement value corresponding to the absolute physical input time as the spatial center point. The processor combines the preset spatial tolerance range to construct a spatial retrieval interval. The processor performs a dual-constraint retrieval operation in the production log of the manufacturing execution system, filtering out production records that simultaneously meet the time retrieval interval conditions and the spatial retrieval interval conditions; the processor extracts a unique batch number code or work group code from the filtered production records as the responsible entity identifier; The processor encapsulates the responsible entity identifier, absolute physical input time, throughput imbalance residual, and traceability confidence into a traceability control data vector; Among them, the source tracing confidence is a quantitative value calculated based on the flux imbalance residual and the global cumulative mixed potential energy. The source tracing confidence is used to numerically measure the credibility of the source tracing results in the physical conservation dimension and the waveform matching dimension.

Images