Automatic potentiometric titration method

Abstract

The invention relates to the field of chemical analysis automation equipment, and discloses an automatic potentiometric titration method which comprises the following steps: before a titration task is started, executing adaptive cleaning and collecting residual spectrum fingerprints of a system; calibrating a titration prediction model in real time by using the fingerprint; performing predictive titration to obtain complete titration data; after the titration is finished, carrying out multi-dimensional analysis on the whole process data, and quantitatively evaluating the pollution risk and the data quality; and finally, dynamically adjusting the cleaning strategy of the next task according to the evaluation result, and adaptively updating the prediction model according to the prediction deviation of the titration. By establishing a closed-loop intelligent working process of cleaning, diagnosis, feedback and learning, the state of the system can be sensed in real time, a cleaning strategy and a prediction model can be dynamically adjusted, the cleaning effect is ensured, self-optimization of the titration process is also realized, the long-term analysis accuracy and the intelligent level of the automatic titration system are improved, and the working efficiency is improved. And the dependence on manual intervention and troubleshooting is reduced.

Description

Technical Field

[0001] This invention relates to the field of automated chemical analysis equipment, specifically an automated potentiometric titration method. Background Technology

[0002] Automated potentiometric titrators, as mature quantitative chemical analysis instruments, have been widely used in pharmaceuticals, chemicals, food safety, environmental monitoring, and other industries. By precisely controlling the addition of titrant and monitoring changes in electrode potential in real time, they automatically determine the endpoint of a chemical reaction, greatly improving the accuracy and reproducibility of analytical results while reducing the workload of operators. Existing automated titrators can effectively execute preset titration procedures and cleaning processes, achieving automation of the analytical process.

[0003] However, current automation technologies primarily remain at the execution level, lacking the ability to perceive and provide feedback on the system's own status and the quality of the analytical process. Current equipment cleaning procedures are typically fixed, such as setting a fixed number of cleaning cycles and time intervals. This open-loop cleaning method cannot confirm whether the system has truly reached a clean state; serious sample residues or cross-contamination are possible, and fixed cleaning procedures are insufficient to completely eliminate interference, leading to biased subsequent analytical results. Conversely, if the sample itself is simple, fixed procedures result in unnecessary reagent consumption and wasted time.

[0004] Furthermore, the system's operating state is not static during titration. Electrode response performance drifts over time, microbubbles exist in the tubing, or unexpected side reactions occur in the sample itself; these factors can all affect the shape of the titration curve and the accuracy of endpoint determination. Current technologies typically lack a mechanism for in-depth analysis of the entire process data and diagnosis of potential problems after the titration task is completed. The system cannot autonomously determine the reliability of the titration results, nor can it dynamically adjust subsequent response strategies based on potential contamination risks, such as strengthening the cleaning efforts for the next task. This lack of diagnostic capability makes problem detection heavily reliant on operator experience, reducing the reliability of automation.

[0005] Furthermore, existing systems, when faced with diverse samples and long-term instrument aging, typically employ static titration algorithms and model parameters that do not self-optimize based on historical performance. Each titration is an independent event, preventing the system from learning and improving from past prediction biases to continuously enhance its titration efficiency and endpoint prediction accuracy for specific sample types. In summary, while current automated titration technologies achieve operational automation, they generally lack an intelligent mechanism capable of assessing system status, diagnosing process quality, and providing closed-loop feedback and adaptive learning. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides an automatic potentiometric titration method. This method solves the problems of existing automatic potentiometric titration methods that typically employ fixed cleaning procedures, which cannot be adjusted according to the actual cleanliness of the pipeline system. This leads to excessive consumption of cleaning agents or incomplete cleaning. Furthermore, the fixed analytical model parameters during the titration process cannot compensate for trace residues in the pipeline or changes in the sample matrix, thus affecting the accuracy of titration endpoint determination.

[0007] To solve the above-mentioned technical problems, the present invention provides an automatic potentiometric titration method, comprising the following steps: S1. Perform the automatic cleaning step and collect the residual spectral fingerprint of the pipeline system after cleaning; S2. Calibrate a titration prediction model based on the residual spectral fingerprint; S3. Perform a predictive titration procedure based on the calibrated titration prediction model; S4. Obtain the titration process data generated by the predictive titration step; S5. Analyze the titration process data to assess the risk of contamination; S6. Based on the assessment results of the pollution risk, determine the cleaning strategy to be used in the next automatic cleaning step.

[0008] Preferably, the automatic cleaning step in step S1 includes: Pump the cleaning agent into the piping system; Real-time monitoring of the spectrum of the outflowing liquid; The spectrum of the outflowing liquid is compared with the preset pure cleaning agent reference spectrum; The cleaning process is terminated when the difference between the spectrum of the outflowing liquid and the preset pure cleaning agent reference spectrum is lower than a preset convergence threshold.

[0009] In one specific embodiment, the residual spectral fingerprinting step of the acquisition pipeline system in step S1 includes: After the automatic cleaning step is completed, pure solvent is pumped into the pipeline system for rinsing. Collect the spectrum of the pure solvent flowing out during the rinsing process; The difference between the spectrum of the outflowing pure solvent and the preset pure solvent reference spectrum is calculated to generate the residual spectral fingerprint.

[0010] The residual spectral fingerprint is calculated as follows: ΔS(λ)=S rinse (λ)-S pure (λ); Wherein, ΔS(λ) represents the spectral difference value as the residual spectral fingerprint; S rinse(λ) represents the spectral intensity of the pure solvent flowing out during the rinsing process; S pure (λ) represents the intensity of the preset pure solvent reference spectrum; λ represents the wavelength of the spectrum.

[0011] Preferably, step S2 includes: Features are extracted from the residual spectral fingerprint; The features include signal noise level, initial potential offset, and matrix effect coefficient; The extracted features are used to calibrate the initial parameters of the titration prediction model to improve the accuracy of titration endpoint prediction in the predictive titration step of step S3.

[0012] In a specific embodiment, step S2 is implemented in the following ways: The standard deviation of the residual spectral fingerprint within a preset flat band without chemical absorption peaks is calculated and used to calibrate the signal-noise level parameters of the titration prediction model. The calibration relationship can be expressed as follows: in, σ represents the calibrated signal-to-noise level parameter. def This represents the default baseline noise level parameter of the device; α represents a preset weighting coefficient; StdDev(·) represents the standard deviation calculation function; The residual spectral fingerprint represents the value within a preset flat band λ. flat The data set within.

[0013] The characteristic absorption peaks are searched and their areas are calculated in the residual spectral fingerprint to calibrate the initial potential shift parameters of the titration prediction model. The calibration relationship can be expressed as follows: ΔE off,calibrated =β·A peak (λ c ); Where, ΔE off,calibrated Represents the initial potential offset parameter after calibration; β represents a preset correlation coefficient; A peak (λ c ) represents the characteristic wavelength λ c The peak area of ​​the absorption peak detected at the location.

[0014] Preferably, step S3, which involves performing a predictive titration, includes: Perform a reconnaissance titration to obtain the corresponding data of potential and volume in the initial stage; The calibrated titration prediction model is fitted using the corresponding potential and volume data from the initial stage to solve for the predicted endpoint volume.

[0015] In one specific embodiment, after determining the predicted endpoint volume, the step of performing the predictive titration further includes: Based on the calculated predicted endpoint volume, the titration path is planned; The titration path includes a rapid approach phase and a fine scan phase; During the rapid approach phase, a large step size is used to add titrant, while during the fine scanning phase, a dynamically adjusted small step size is used to add titrant.

[0016] Preferably, step S5 includes: Analyze the signal noise level of the titration process data; Analyze the curve shape of the titration process data; Analyze the reaction kinetic parameters of the titration process data; When the signal noise level is higher than a preset value, the curve shape is distorted, or the reaction kinetic parameters show a slow reaction, the pollution risk is assessed as high risk.

[0017] In one specific embodiment, step S6 includes: When the pollution risk assessment is low, a standard cleaning strategy is determined to be adopted. When the pollution risk assessment is high, an enhanced cleaning strategy is determined to be adopted; The enhanced cleaning strategies include selecting different types of cleaning agents, increasing the cleaning temperature, or increasing the cleaning pressure.

[0018] Preferably, after performing the predictive titration step, the method further includes the following steps: Calculate the actual final volume; The internal coefficients of the titration prediction model are updated using the deviation between the predicted endpoint volume and the actual endpoint volume to optimize the predictive performance of the titration prediction model in subsequent titration tasks. The updating of the internal coefficients can be performed using the following relationship: Where, Θ new Represents the set of internal coefficients of the updated titration prediction model; Θ old η represents the set of internal coefficients of the titration prediction model before the update; η represents the learning rate used to control the update step size. L(V ep V ea ) represents the volume V based on the predicted endpoint. ep With respect to the actual endpoint volume V ea The calculated loss function value; This represents the gradient operator of the loss function L with respect to the internal coefficient set Θ.

[0019] This invention provides an automatic potentiometric titration method, which has the following beneficial effects: 1. This invention collects the residual spectral fingerprint of the pipeline system after automatic cleaning and uses this fingerprint to calibrate the titration prediction model. This allows the prediction model to pre-compensate for the baseline drift and noise level of the potential signal caused by trace residues or matrix effects in the pipeline system before the titration begins. This enables the subsequent predictive titration to obtain a more accurate titration endpoint prediction value, thereby improving the accuracy of automatic potentiometric titration analysis.

[0020] 2. This invention analyzes titration process data to assess contamination risk and determines the strategy for subsequent cleaning operations based on the assessment results. It establishes a feedback mechanism that transforms the cleaning operation from a fixed procedure into an adaptive process that is dynamically adjusted according to actual analytical conditions. This ensures that appropriate cleaning intensities are applied to samples with different levels of contamination, avoiding insufficient cleaning or excessive consumption of cleaning agents. At the same time, it reduces the need for manual intervention and improves the automation level and overall efficiency of equipment operation.

[0021] 3. This invention updates the internal coefficients of the titration prediction model by utilizing the deviation between the predicted endpoint volume and the actual endpoint volume. This iterative update based on historical performance, combined with real-time calibration based on residual spectral fingerprints, enables the titration system to adapt not only to instantaneous state changes before a single measurement, but also to systematic drift caused by electrode aging, tubing wear, or long-term changes in sample batches. This enhances long-term operational stability and adaptability to different operating conditions. Attached Figure Description

[0022] Figure 1 This is a structural block diagram of an automatic potentiometric titration system according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating the automatic cleaning and residual spectral fingerprint acquisition steps according to an embodiment of the present invention. Figure 3 This is a schematic flowchart illustrating the calibration steps of a titration prediction model according to an embodiment of the present invention. Figure 4 This is a flowchart illustrating a predictive titration step according to an embodiment of the present invention. Figure 5 This is a flowchart illustrating the data analysis and pollution risk assessment steps of one embodiment of the present invention; Figure 6 This is a flowchart illustrating the steps for determining and adjusting a cleaning strategy according to an embodiment of the present invention. Figure 7 This is a flowchart illustrating the adaptive update steps of a titration prediction model according to an embodiment of the present invention. Figure 8 This is a schematic diagram of the internal structure of the device of the present invention; Figure 9 This is a three-dimensional schematic diagram of the device of the present invention.

[0023] Legend 10. Central control unit; 20. Sample processing unit; 30. Titration unit; 40. Cleaning unit; 50. Spectroscopic detection unit; 60. Main body of the equipment; 201. Sample cup; 202. Stirring table; 203. Electrode holder; 204. Titrator; 70. Desiccant; 301. Clamping head; 401. Liquid inlet cap; 402. Filter; 403. Micron filter element; 404. Manual valve stem. Detailed Implementation

[0024] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] See attached document Figure 1 , Figure 1 This is a structural block diagram of an automated potentiometric titration system according to an embodiment of the present invention. The present invention provides an automated potentiometric titration method, which can be implemented on an automated potentiometric titration system.

[0026] The system may include: a central control unit 10, a sample processing unit 20, a titration unit 30, a cleaning unit 40, and a spectral detection unit 50.

[0027] The central control unit 10, such as a microcontroller or industrial computer with an integrated processor and memory, is electrically connected to other units of the system via an internal bus. The central control unit 10 is equipped with a touch screen as a human-machine interface and is used to execute preset control programs and algorithms.

[0028] The sample processing unit 20 includes an automatic sample stage for carrying and transporting multiple sample cups, and an electrode holder for fixing the measuring electrode, reference electrode, and burette outlet. The measuring electrode is electrically connected to the central control unit 10 for transmitting potential signals to the central control unit 10.

[0029] The titration unit 30 includes one or more titration units and titration reagent bottles connected thereto. Under the command of the central control unit 10, the titration unit 30 precisely pumps the titration reagent into the sample cup of the sample processing unit 20 through tubing.

[0030] The cleaning unit 40 includes a separate pump and piping system, or reuses the pump and piping of the titration unit 30. The cleaning unit 40 is used to connect to an inlet bottle containing cleaning agent or pure solvent and to perform cleaning and rinsing operations on the piping system.

[0031] The spectral detection unit 50 is installed in the waste liquid discharge pipeline of the system and includes a light source and a spectral sensor. The spectral detection unit 50 is used to measure the spectral data of the liquid flowing through the pipeline and transmit the data to the central control unit 10 in real time.

[0032] In one embodiment of the present invention, the overall process of the automatic potentiometric titration method is as follows: Before performing the titration analysis, the central control unit 10 controls the cleaning unit 40 to perform an automatic cleaning step, and after cleaning, controls it to pump in pure solvent for rinsing. During the rinsing process, the spectral detection unit 50 collects the spectrum S of the outflowing solvent in real time. rinse (λ), and transmit it to the central control unit 10.

[0033] The central control unit 10 calculates the spectrum and compares it with the pure solvent reference spectrum S pre-stored in its memory. pure The difference between (λ) is used to generate the residual spectral fingerprint ΔS(λ).

[0034] ΔS(λ)=S rinse (λ)-S pure (λ); Where ΔS(λ) represents the spectral difference as a residual spectral fingerprint, and is a function of the spectral wavelength λ; S rinse (λ) represents the spectral intensity of the pure solvent flowing out during the rinsing process; S pure (λ) represents the preset pure solvent reference spectral intensity.

[0035] Subsequently, the central control unit 10 performs parameter calibration on a titration prediction model stored within it, based on the generated residual spectral fingerprint ΔS(λ). The parameter calibration process includes extracting features from the fingerprint and using them to correct the model's initial parameters, such as the signal-noise level parameter and the initial potential offset parameter.

[0036] After model calibration, the central control unit 10 controls the sample processing unit 20 to position the sample cup to be tested below the electrode holder, and controls the titration unit 30 to perform a predictive titration step. The predictive titration step includes performing a reconnaissance titration to quickly solve for the predicted endpoint volume, and planning and executing a titration path based on the predicted volume, which includes a rapid approach phase and a fine scan phase.

[0037] After the predictive titration is completed, the central control unit 10 acquires and analyzes the complete titration process data to assess the contamination risk. Based on the assessed contamination risk level, the central control unit 10 then determines and sets the cleaning strategy for the next automatic cleaning step, such as selecting a standard cleaning strategy or an enhanced cleaning strategy, thereby forming a closed-loop control.

[0038] In addition, the central control unit 10 calculates the actual endpoint volume of this titration and compares it with the predicted endpoint volume. The deviation value is used to adjust the internal coefficients of the titration prediction model according to the preset update algorithm, so as to optimize its prediction performance in subsequent titration tasks.

[0039] See attached document Figure 2 , Figure 2 This is a flowchart illustrating the automatic cleaning and residual spectral fingerprint acquisition steps according to an embodiment of the present invention. This section describes in detail the implementation of the first stage of the method of the present invention, namely step S1.

[0040] At the start of a titration task, the central control unit 10 first invokes the cleaning unit 40 and the spectral detection unit 50 to execute an adaptive automatic cleaning process. The central control unit 10 instructs the pump of the cleaning unit 40 to pump cleaning agent from the inlet bottle containing the cleaning agent. The cleaning agent flows sequentially through the system's piping, valves, and the outlet of the electrode holder. During this period, the spectral detection unit 50, installed in the waste liquid line, is continuously activated, acquiring the real-time spectrum S of the outflowing liquid at preset time intervals (e.g., 10 times per second). outflow (λ), and send the data to the central control unit 10.

[0041] The central control unit 10 receives the real-time spectrum S outflow After (λ), it is compared with the pre-stored pure cleaning agent reference spectrum S in the memory. baseline (λ) is compared to calculate a quantified measure of cleaning difference D. c In a specific implementation, this metric D c It is obtained by calculating the absolute difference of the integral of the two spectra over the specified monitoring wavelength range [λ1, λ2]. in: D c S represents the calculated cleaning difference metric. outflow (λ) represents the real-time spectral intensity of the effluent collected by the spectral detection unit 50; S baseline(λ) represents the reference spectral intensity of the pre-calibrated pure cleaning agent; λ1, λ2 represent integral signs, indicating an accumulation process, which means continuously summing the values ​​after the signs within the specified wavelength range, that is, the interval from wavelength λ1 to λ2; dλ represents wavelength differentiation, indicating that the integration is carried out along the wavelength λ axis.

[0042] The central control unit 10 will calculate the difference measure D in real time. c With a preset convergence threshold T c Compare the convergence threshold T. c This is a parameter pre-set according to the experimental accuracy requirements and stored in the configuration of the central control unit 10. If D c Greater than or equal to T c If the central control unit 10 determines that the cleaning is not complete, it will continue to instruct the cleaning unit 40 to pump in cleaning agent. If D c Less than T c If the central control unit 10 determines that the pipeline has reached the preset cleanliness standard, it will immediately send a command to terminate the pumping of the cleaning agent.

[0043] After the automatic cleaning step is terminated, in order to generate a residual spectral fingerprint, the central control unit 10 immediately instructs the cleaning unit 40 to switch the flow path and pump pure solvent (e.g., the background solvent used in the titration reaction) from the inlet bottle to perform a rinse on the entire pipeline system. The purpose of this rinse step is to wash away the cleaning agent remaining in the pipeline and dissolve or carry away any trace residues adsorbed on the inner wall of the pipeline that were not completely removed in the aforementioned cleaning step.

[0044] During the rinsing process, the spectral detection unit 50 is also activated to acquire the spectrum of the outflowing pure solvent. To improve the signal-to-noise ratio, the central control unit 10 can acquire spectral data over a continuous period of time (e.g., 2 seconds) after detecting a stable flow rate, and then average the data to obtain a stable rinsing spectrum S. rinse (λ).

[0045] Finally, the central control unit 10 retrieves from its memory the pre-stored pure solvent reference spectrum S corresponding to the rinsing solvent. pure (λ), and perform a subtraction operation to generate a data array as the residual spectral fingerprint ΔS(λ) of this measurement: ΔS(λ)=S rinse (λ)-S pure (λ); Where ΔS(λ) represents the spectral difference as a residual spectral fingerprint; S rinse (λ) represents the spectral intensity of the pure solvent flowing out during the rinsing process; S pure (λ) represents the intensity of the preset pure solvent reference spectrum; λ represents the wavelength of the spectrum.

[0046] The generated residual spectral fingerprint ΔS(λ) is stored in the central control unit 10 and used as a direct input for the titration prediction model calibration in the subsequent step S2. Through the above process, the physical cleanliness state of the pipeline system is converted into a quantitative digital signal that can be directly processed by the algorithm.

[0047] See attached document Figure 3 , Figure 3 This is a schematic flowchart of the titration prediction model calibration steps according to an embodiment of the present invention. After obtaining the residual spectral fingerprint ΔS(λ) generated in the previous stage, the central control unit 10 immediately executes step S2 to calibrate the parameters of the titration prediction model stored therein.

[0048] The central control unit 10 first analyzes the residual spectral fingerprint ΔS(λ) in the form of a data array to extract at least two types of key features. The first type of feature is the signal noise level reflecting the physical noise of the system, and the second type of feature is the initial potential shift reflecting the influence of specific chemical residues.

[0049] To extract the signal noise level, the processor of the central control unit 10 operates in a preset spectrally flat band λ that does not contain the characteristic absorption peaks of known chemical substances. flat Within this model, the standard deviation of the residual spectral fingerprint is calculated. This value is used to calibrate the signal noise level parameter in the titration prediction model, which describes the random fluctuations of the potential signal. Its calibration formula is: in, σ represents the calibrated signal-to-noise level parameter. def α represents the default parameter stored in the central control unit 10 that characterizes the device's own basic electronic noise; α represents a preset weighting coefficient used to quantify the correlation between spectral noise and potential noise; StdDev(·) represents the standard deviation calculation function. Represents the residual spectral fingerprint in the flat band λ flat The set of data points within.

[0050] To extract the initial potential offset, the processor of the central control unit 10 searches for the characteristic absorption wavelength λ of known residues in a peak feature library stored internally. c The processor matches the residual spectral fingerprint ΔS(λ) with the feature library, searching for absorption peaks near preset wavelength positions. If a peak is detected, its peak area A is calculated by numerically integrating the region under the peak profile. peak (λ c The peak area was used to calibrate the initial potential shift parameter ΔE in the titration prediction model. off,calibrated The calibration formula is: ΔE off,calibrated =β·A peak (λ c ); Where, ΔE off,calibrated Represents the initial potential shift parameter after calibration; β represents a correlation coefficient preset for a specific residue, characterizing the potential shift caused by a unit peak area; A peak (λ c ) represents the characteristic wavelength λ c The peak area of ​​the absorption peak detected at the location.

[0051] The titration prediction model is a parameterized mathematical function E = f(V, Θ) that describes the relationship between potential E and titrant volume V, where Θ is the set of parameters of the model. The calibration process specifically involves the central control unit 10 converting the above-calculated... and ΔE off,calibrated This is used to update the initial parameters of the model for this titration task. For example, the initial potential parameters of the model are set to the theoretical values ​​and ΔE. off,calibrated The sum. Simultaneously, the calibrated noise parameters. This will be used as a weighting factor in the subsequent curve fitting algorithm to reduce the impact of high-noise data points on the fitting results. Through this step, the titration prediction model already includes compensation information for the current physical and chemical state of the pipeline system before performing predictions.

[0052] See attached document Figure 4 , Figure 4 This is a schematic flowchart of a predictive titration step according to an embodiment of the present invention. After the titration prediction model has been calibrated, the central control unit 10 then initiates step S3 to perform the predictive titration.

[0053] First, the central control unit 10 instructs the sample processing unit 20 to move the sample cup containing the sample to be tested to the area below the electrode holder, and to immerse the measuring electrode, reference electrode, and tibial tube outlet into the sample solution. Then, the central control unit 10 begins the reconnaissance titration. During this stage, the central control unit 10 instructs the titration unit 30 to add 2 to 4 steps of titrant to the sample cup in preset, relatively large volume increments (e.g., 5-10% of the expected endpoint volume). After each step is completed, the central control unit 10 records the stabilized potential value, thereby obtaining a set of initial potential-volume correspondence data points (Vt). i E i ).

[0054] After acquiring this set of initial data points, the processor of the central control unit 10 immediately invokes a nonlinear least squares fitting algorithm. This algorithm uses these initial data points to fit the titration prediction model E = f(V, Θ) calibrated in step S2. calibratedThe processor performs a fitting operation. After fitting, it solves for the points where the second derivative of the fitted curve is zero (i.e., d). 2 E / dV 2 =0) to determine the inflection point of the curve, and the volume corresponding to this inflection point is determined as the predicted endpoint volume V. ep,predicted .

[0055] Based on the calculated predicted endpoint volume V ep,predicted The central control unit 10 plans the subsequent titration path, which is divided into a rapid approach phase and a fine scan phase. The endpoint volume of the rapid approach phase, i.e., the approach volume V, is... near It is set as a specific scale of the predicted endpoint volume, such as V. near =0.9×V ep,predicted The fine scanning stage covers from V... near To a region exceeding the predicted endpoint volume, for example, 1.1 × V ep,predicted .

[0056] After path planning is completed, the central control unit 10 instructs the titration unit 30 to begin executing the path. During the rapid approach phase, the titration unit 30 continuously adds titrant to the sample cup at its maximum pump speed or in very large step sizes until the cumulative added volume reaches the approach volume V. near At this stage, the frequency of potential monitoring can be reduced to focus on rapid liquid addition.

[0057] When the titration volume reaches V near Subsequently, the central control unit 10 automatically switches to the fine scanning stage. In this stage, the titrant step size ΔV added at each step is dynamically calculated and adjusted. Before adding the titrant in the i-th step, the central control unit 10 first uses a calibrated prediction model f(V,Θ) to... calibrated Calculate the predicted slope of the titration curve at the current volume. Next titration step size ΔV i Then it is determined according to the following relationship: Where, ΔV i This represents the volume of titrant to be added in the next step; k is a preset proportionality coefficient used to adjust the step size sensitivity; ΔV min The minimum step size is determined by the mechanical precision of the titration unit 30; ΔV max This is the maximum allowable step size set for this stage to prevent excessively large step sizes in areas where the curve is particularly flat.

[0058] In this way, in the flat region far from the endpoint, the predicted slope is small, and the titration step size is large; as the endpoint abruptly approaches, the predicted slope increases sharply, and the titration step size automatically decreases to the minimum step size level. The central control unit 10 cyclically executes the operations of predicting the slope, calculating the step size, adding titrant, and measuring potential until the titration volume exceeds the preset upper limit of the fine scan stage, thereby accurately capturing the complete curve near the titration endpoint with high data density.

[0059] See attached document Figure 5 , Figure 5 This is a flowchart illustrating the data analysis and pollution risk assessment steps according to an embodiment of the present invention. After the predictive titration in step S3 is completed, the central control unit 10 immediately proceeds to steps S4 and S5 to analyze and evaluate the status of the titration process.

[0060] First, in step S4, the central control unit 10 integrates the data recorded and stored throughout the titration process into a structured titration process dataset. The titration process dataset includes not only a complete set of corresponding potential and volume data points {(V i E i It also includes the set of time {Δt} required for the potential signal to reach a preset stability standard after each titrant addition step. stable,i}

[0061] Subsequently, in step S5, the processor of the central control unit 10 performs a multi-dimensional quantitative analysis of the titration process dataset to assess the contamination risk. The quantitative analysis process includes the following calculations: The signal-noise level of the titration process data is analyzed. The processor first automatically identifies two potential flat regions before and after the stoichiometric point based on the overall shape of the titration curve. Then, the potential data points E within these two regions are calculated respectively. i The standard deviation of the two standard deviations is taken as the larger of the two standard deviations or their arithmetic mean, and denoted as N, which measures the signal noise level of this titration.

[0062] The curve shape of the titration process data is analyzed. The processor analyzes the corresponding data of potential and volume {(V i E i Numerical differentiation was performed to calculate the first derivative (dE / dV) and second derivative (dV / dV) of the titration curve. 2 E / dV 2 The processor locates the main peak on the second derivative curve and calculates the ratio of the left width to the right width of the main peak at half the peak height, thus obtaining a peak shape symmetry coefficient A. s Simultaneously, the processor executes a peak search algorithm on the second derivative curve to count the number P of secondary peaks, excluding the main peak, whose amplitudes exceed a preset noise threshold. sec.

[0063] The reaction kinetic parameters of the titration process data are analyzed. The processor analyzes the collected set of steady-state times {Δt}. stable,i In the formula, the arithmetic mean T of all stationary times is calculated. avg_stable Especially the average stabilization time near the titration endpoint jump zone.

[0064] The processor of the central control unit 10, based on a preset rule set, processes the calculated quantitative indicators (N, A) s, P sec, T avg_stable ) and a set of thresholds (T) stored in its configuration N ,T A1 ,T A2 ,T P ,T T The logic of this rule set is as follows: If the signal-noise level metric N is higher than the threshold T N Or the peak shape symmetry coefficient A s, Less than the lower threshold T A1 or greater than the upper limit threshold T A2 or the number of secondary peaks P sec Greater than threshold T P or mean settling time T avg_stable Greater than threshold T T At that time, the central control unit 10 sets the pollution risk assessment result of this titration as a discrete value: high risk.

[0065] If none of the above conditions are met, the central control unit 10 will set the pollution risk assessment result to another discrete value: low risk. This assessment result will be stored and used as the direct input for determining the cleaning strategy in step S6.

[0066] See attached document Figure 6 , Figure 6 This is a flowchart illustrating the steps for determining and adjusting a cleaning strategy according to an embodiment of the present invention. After completing the contamination risk assessment in step S5, the central control unit 10 immediately executes step S6 to determine the cleaning strategy to be adopted at the start of the next titration task based on the assessment results.

[0067] The central control unit 10 has at least two cleaning strategies pre-configured in its memory: a standard cleaning strategy and an enhanced cleaning strategy.

[0068] A standard cleaning strategy is defined as a fixed baseline cleaning procedure. For example, the parameter set of this strategy may specify: using a first cleaning agent (e.g., deionized water), flushing the piping system for 30 seconds at a constant flow rate at room temperature.

[0069] Enhanced cleaning strategies are defined as combinations of one or more enhanced cleaning operations. These operations are stored in the central control unit 10 and can be invoked according to preset rules.

[0070] The specific operation involves selecting a second or third cleaning agent that differs from the cleaning agent used in the standard cleaning strategy. This operation is controlled by the central control unit 10, which controls the flow path selection valve of the cleaning unit 40 to pump cleaning agents with specific chemical properties (e.g., acidic solutions, alkaline solutions, or organic solvents) from different inlet bottles.

[0071] Increase the temperature during the cleaning process. If the system is equipped with a heating module, the central control unit 10 sends a control signal to the module to heat the cleaning agent flowing through the pipeline to a preset temperature value (e.g., 60°C) and maintain that temperature for a preset period of time.

[0072] Increase the physical scouring intensity of the cleaning process. The central control unit 10 adjusts the drive signal sent to the pump in the cleaning unit 40 to make it operate at a flow rate higher than the standard flow rate, or in a pulse mode with a preset frequency and duty cycle, thereby generating turbulence or hydraulic pulses in the pipeline.

[0073] The central control unit 10 executes a conditional judgment logic. It reads the pollution risk assessment results stored in step S5.

[0074] If the assessment result is low risk, the central control unit 10 sets in its internal task queue that the standard cleaning strategy will be invoked and executed when the next analysis cycle begins.

[0075] If the assessment result is high-risk, the central control unit 10 invokes and executes an enhanced cleaning strategy. In one specific embodiment, the central control unit 10 executes a preset sequence of multiple enhanced cleaning operations. For example, rinsing first with an acidic cleaning agent, then with deionized water, and finally with an organic solvent to deal with complex mixed residues. The parameters of this enhanced cleaning strategy (such as cleaning agent type, temperature, and time) are also preset and stored in the central control unit 10.

[0076] This step establishes a closed-loop feedback control system. The analysis results of the previous titration directly determine the strength of the system preparation measures to be taken before the next titration. This decision-making and execution process is entirely automated by the central control unit 10, requiring no manual intervention.

[0077] See attached document Figure 7 , Figure 7This is a flowchart illustrating the adaptive update steps of a titration prediction model according to an embodiment of the present invention. In addition to immediate calibration and feedback control for a single titration, the method of the present invention also includes a long-term adaptive update mechanism for the titration prediction model.

[0078] After each titration task is completely completed, the processor of the central control unit 10 first processes the complete titration process data {(V} obtained in step S4. i E i The final analysis is then performed to calculate the actual endpoint volume V of this titration. ea This calculation is performed using one or more pre-defined standard endpoint calculation methods, such as calculating the volume corresponding to the maximum value of the second derivative of the titration curve, or performing linear extrapolation using the Grignard plot method. The calculated V ea It is considered the true reference value for this titration.

[0079] Subsequently, the central control unit 10 retrieves the predicted endpoint volume V calculated and used in step S3 of this titration from its internal memory. ep The processor calculates V. ep With respect to the actual endpoint volume V ea The deviation between the two is used as an error signal to drive the update of the internal coefficients of the titration prediction model.

[0080] This update process is implemented using a gradient descent-based algorithm. The central control unit 10 performs the following calculations to generate a new set of internal model coefficients Θ. new : Where, Θ new This represents the updated set of internal coefficients of the titration prediction model stored in the central control unit 10 for subsequent titration tasks; Θ old L(V) represents the set of internal coefficients used by the model before this titration begins; η represents the learning rate, a preset positive scalar parameter used to control the magnitude of each update; L(V) represents the learning rate. ep V ea ) represents the volume V based on the predicted endpoint. ep With respect to the actual endpoint volume V ea The calculated loss function value. In a specific implementation, this loss function is defined as the square of the prediction error, i.e., L = (V ep -V ea ) 2 ; This represents the gradient operator, used to calculate the partial derivative of the loss function L with respect to each coefficient in the internal coefficient set Θ.

[0081] The processor of the central control unit 10 calculates the gradient vector. Then, perform vector operations on the above formula to obtain a new set of coefficients Θ. new .

[0082] Finally, the central control unit 10 converts the set of internal coefficients in its memory used for the titration prediction model into Θ. old Updated to Θ new This new set of coefficients Θ new These parameters will serve as the baseline model parameters before performing step S2 calibration in the next titration task. This update mechanism allows the model to be adjusted based on historical predicted performance to compensate for systematic drift caused by electrode aging, reagent batch differences, or slow changes in environmental conditions.

[0083] See attached document Figure 8 -Appendix Figure 9 The system configuration used in this embodiment is as follows: Central control unit 10: Integrated inside the main body of the equipment 60, the operator issues instructions, monitors the process and views the final report through the touch screen on the front of the equipment.

[0084] Sample processing unit 20: consists of a stirring table 202 for placing sample cup 201 and an electrode holder 203 for fixing pH composite electrode.

[0085] Titration unit 30: Its core is a precision piston burette. The titrant (in this example, a NaOH standard solution) is stored in the titrant 204. The mouth of the reagent bottle is filled with desiccant 70 to prevent deliquescence. The titrant is added to the sample cup 201 through a pointed tip.

[0086] Cleaning unit 40: This is a multi-channel flow path system. The cleaning fluid or pure solvent enters the system through a pipe equipped with an inlet cap 401. Before entering the main line, the liquid must flow through a filter 402 with a built-in micron filter element 403 to remove potential particulate matter. A manual valve stem 404 on the pipeline is used for maintenance or emergency evacuation. A clamping head 301 is located below the manual valve stem 404.

[0087] Spectral detection unit 50: A miniature flow cell spectrometer installed on the waste liquid discharge pipeline for real-time monitoring of the spectral characteristics of the outflowing liquid.

[0088] Implementation process: Preparation: The operator weighs a certain amount of industrial hydrochloric acid sample, dilutes it, and places it in sample cup 201. The sample cup is then placed on the stirring table 202, and the electrode is immersed in the solution through the electrode holder 203. Subsequently, the operator selects the hydrochloric acid content determination method on the touch screen and clicks "Start".

[0089] Adaptive cleaning and fingerprint acquisition: The central control unit 10 first executes the cleaning strategy determined in the previous task. It commands the cleaning unit 40 to pump in cleaning agent, which is then purified by filter 402 and micron filter 403, and then rinses the entire burette, nozzle, and electrode surface.

[0090] The spectral detection unit 50 monitors the waste liquid spectrum in real time. The central control unit 10 continuously calculates the cleaning difference metric Dc until the value is lower than a preset threshold, confirming that the system is clean.

[0091] Subsequently, pure solvent was pumped into the tubing to rinse it, and the residual spectral fingerprint at this moment was collected for use in subsequent steps.

[0092] Model calibration: The central control unit 10 analyzed the residual spectral fingerprint and found a small impurity peak in the fingerprint. Based on this, it was determined that the system may have a slight potential response delay. It automatically fine-tuned the electrode response time constant in the prediction model to compensate for this potential effect.

[0093] Predictive titration: The titration unit 30 extracts the solution from the titrator 204 through the nozzle and quickly adds a few drops of NaOH standard solution to the sample cup 201 to complete the reconnaissance titration.

[0094] Based on the initial data points and the calibrated model, the central control unit 10 quickly calculates the predicted endpoint volume (e.g., 15.8 mL) and plans the most efficient titration path.

[0095] The equipment performs a titration process that rapidly approaches a fine scan, with the stirring table 202 working continuously to ensure uniform reaction. The entire titration curve is displayed in real time on the touch screen.

[0096] Pollution risk assessment: After the titration was completed, the central control unit 10 integrated all the data for analysis and found that a slight tailing phenomenon appeared in the latter half of the first derivative peak of the titration curve, which indicated that the reaction kinetic parameters were abnormal.

[0097] The evaluation results pop up on the touchscreen: High risk to abnormal curve shape, possibly due to interference from weak acidic impurities.

[0098] Closed-loop feedback and model update: Cleaning strategy adjustment: Due to the high risk assessment result, the central control unit 10 automatically adjusts the cleaning strategy for the next task to enhanced cleaning, adding an extra alkaline cleaning step to remove any possible acidic residues.

[0099] Model Adaptive Update: Simultaneously, the central control unit 10 analyzes the complete actual titration curve using rigorous mathematical methods, calculating the actual endpoint volume of this titration to be 15.9 mL. It compares this value with the predicted 15.8 mL, calculates a prediction deviation of +0.1 mL, and uses this deviation to update the internal coefficients of the prediction model through a gradient descent algorithm, enabling the model to make more accurate predictions when dealing with similar samples in the future.

Claims

1. An automatic potentiometric titration method, characterized in that, Includes the following steps: S1. Perform the automatic cleaning step and collect the residual spectral fingerprint of the pipeline system after cleaning; S2. Calibrate a titration prediction model based on the residual spectral fingerprint; S3. Perform a predictive titration procedure based on the calibrated titration prediction model; S4. Obtain the titration process data generated by the predictive titration step; S5. Analyze the titration process data to assess the risk of contamination; S6. Based on the assessment results of the pollution risk, determine the cleaning strategy to be used in the next automatic cleaning step.

2. The automatic potentiometric titration method according to claim 1, characterized in that, The automatic cleaning step in step S1 includes: Pump the cleaning agent into the piping system; Real-time monitoring of the spectrum of the outflowing liquid; The spectrum of the outflowing liquid is compared with the preset pure cleaning agent reference spectrum; The cleaning process is terminated when the difference between the spectrum of the outflowing liquid and the preset pure cleaning agent reference spectrum is lower than a preset convergence threshold.

3. The automatic potentiometric titration method according to claim 1, characterized in that, The residual spectral fingerprinting step of the acquisition pipeline system in step S1 includes: After the automatic cleaning step is completed, pure solvent is pumped into the pipeline system for rinsing. Collect the spectrum of the pure solvent flowing out during the rinsing process; The difference between the spectrum of the outflowing pure solvent and the preset pure solvent reference spectrum is calculated to generate the residual spectral fingerprint.

4. The automatic potentiometric titration method according to claim 1, characterized in that, Step S2 includes: Features are extracted from the residual spectral fingerprint; The features include signal noise level, initial potential offset, and matrix effect coefficient; The extracted features are used to calibrate the initial parameters of the titration prediction model to improve the accuracy of titration endpoint prediction in the predictive titration step of step S3.

5. The automatic potentiometric titration method according to claim 1, characterized in that, Step S3, which involves performing a predictive titration, includes: Perform a reconnaissance titration to obtain the corresponding data of potential and volume in the initial stage; The calibrated titration prediction model is fitted using the corresponding potential and volume data from the initial stage to solve for the predicted endpoint volume.

6. The automatic potentiometric titration method according to claim 5, characterized in that, The predictive titration step further includes: Based on the calculated predicted endpoint volume, the titration path is planned; The titration path includes a rapid approach phase and a fine scan phase; During the rapid approach phase, a large step size is used to add titrant, while during the fine scanning phase, a dynamically adjusted small step size is used to add titrant.

7. The automatic potentiometric titration method according to claim 1, characterized in that, Step S5 includes: Analyze the signal noise level of the titration process data; Analyze the curve shape of the titration process data; Analyze the reaction kinetic parameters of the titration process data; When the signal noise level is higher than a preset value, the curve shape is distorted, or the reaction kinetic parameters show a slow reaction, the pollution risk is assessed as high risk.

8. The automatic potentiometric titration method according to claim 7, characterized in that, Step S6 includes: When the pollution risk assessment is low, a standard cleaning strategy is determined to be adopted. When the pollution risk assessment is high, an enhanced cleaning strategy is determined to be adopted; The enhanced cleaning strategies include selecting different types of cleaning agents, increasing the cleaning temperature, or increasing the cleaning pressure.

9. The automatic potentiometric titration method according to claim 5, characterized in that, After performing the predictive titration step, the following steps are also included: Calculate the actual final volume; The internal coefficients of the titration prediction model are updated using the deviation between the predicted endpoint volume and the actual endpoint volume to optimize the prediction performance of the titration prediction model in subsequent titration tasks.

10. The automatic potentiometric titration method according to claim 1, characterized in that, Step S2 includes: Calculate the standard deviation of the residual spectral fingerprint in a specified band, which is used to calibrate the signal-noise level parameters of the titration prediction model; The signal noise level parameter is used to model the fluctuations of the potential signal in the predictive titration step; Search for characteristic absorption peaks and calculate peak areas in the residual spectral fingerprint to calibrate the initial potential shift parameters of the titration prediction model; The initial potential offset parameter is used to compensate for the initial offset of the potential measurement in the predictive titration step.

Images