Multi-mode liquid level detection method and system for pipetting operation

Abstract

The invention relates to a multi-mode liquid level detection method and system for pipetting operation, and belongs to the technical field of laboratory automatic liquid treatment. The method comprises the following steps: acquiring pipetting operation detection parameters, and screening detection distribution points in a pipetting stroke domain through a time sequence change point detection algorithm to obtain a pipetting operation feature sequence; constructing a pipetting response sensing matrix, calculating an equipment state monitoring weight corresponding to the detection mode according to the response correlation value, and generating an equipment state monitoring scheme; the liquid level detection process is modeled, quantiles of the pipetting detection equipment in different operation states are calculated through a quantile statistics evaluation algorithm, and a liquid level detection result is output; according to the liquid level detection result, abnormal operation offset parameters are adjusted based on a response feedback consistency mechanism, liquid level detection signals are mapped to a pipetting operation-response state chain, contribution coefficients of all detection modes are corrected, and an equipment state monitoring scheme is updated.

Description

Technical Field

[0001] This invention belongs to the field of laboratory automated liquid handling technology, specifically relating to a multimodal liquid level detection method and system for pipetting operations. Background Technology

[0002] Liquid pipetting is a core step in laboratory liquid handling, and its accuracy directly affects the reliability and repeatability of experimental results. Current technologies primarily rely on single-modal sensors, such as optical or capacitive sensors, to monitor the liquid surface position in real time. However, single-modal detection has several limitations: optical methods are significantly affected by liquid color, transparency, and ambient light, making them unsuitable for complex experimental conditions; capacitive or pressure sensors are sensitive to liquid vibration and bubble interference, easily generating noise signals and leading to inaccurate liquid level judgments. Furthermore, traditional liquid level detection methods typically employ fixed thresholds or simple filtering algorithms, lacking effective analysis of the temporal changes and stage-specific characteristics of the liquid level response signal, making it difficult to distinguish between genuine liquid level changes and abnormal responses caused by equipment status fluctuations.

[0003] During pipetting operations, minute changes in liquid volume and the discontinuity of pipetting operation lead to non-stationary and time-varying characteristics in the liquid surface signal, increasing the complexity of liquid surface detection. Existing methods struggle to achieve detailed modeling of the operational phases and lack hierarchical response analysis tools for different operational phases. In particular, regarding multimodal signal fusion, traditional techniques lack effective weighting strategies and cannot fully utilize the complementary information from multiple sources such as optical, capacitive, and pressure signals, thus affecting the accuracy and stability of liquid surface detection.

[0004] Meanwhile, fluctuations in equipment condition, such as pipette vibration, mechanical wear, or environmental disturbances, can introduce transient interference, further reducing detection reliability. Most existing technologies do not consider the dynamic coupling relationship between liquid surface response and equipment condition, and lack adaptive correction mechanisms based on response feedback. This leads to anomalies in liquid surface detection results, making it difficult to meet the experimental requirements of high precision and high reliability. Furthermore, existing systems typically only provide a single numerical value or signal when outputting liquid surface detection results, lacking staged response mapping and continuity constraints, thus failing to provide usable closed-loop information for automated control or equipment condition monitoring.

[0005] Therefore, how to achieve high-precision, multimodal, and phased liquid level detection during pipetting operations, combine it with equipment status monitoring to correct abnormal responses, and generate continuous and reliable liquid level detection results are key issues that laboratory liquid handling technology urgently needs to address. Summary of the Invention

[0006] To address the aforementioned problems in the prior art, this invention provides a multimodal liquid level detection method for pipetting operations. The objective of this invention can be achieved through the following technical solutions: S1: Obtain the detection parameters of the pipetting operation, identify candidate liquid surface response points through the time-series change point detection algorithm, generate liquid surface detection signals, construct a liquid surface pipetting signal window based on the liquid surface detection signals, introduce a multi-modal detection attenuation factor to calculate the response correlation value corresponding to the current pipetting operation, and obtain the pipetting operation feature sequence; S2: Construct a pipetting response sensing matrix, obtain the temporal evolution relationship of the pipetting operation feature sequence under different pipetting stages, and query the equipment status monitoring weight corresponding to the current pipetting stage based on the response correlation value to generate an equipment status monitoring scheme; S3: Execute the equipment status monitoring scheme, model the liquid level detection process, calculate the quantiles of the liquid level detection device under different operating states through the quantile statistical evaluation algorithm, and perform state-sensitive correction on the normal liquid level response and abnormal response caused by equipment status fluctuations in the liquid level detection signal based on the quantiles, and output the liquid level detection results. S4: Adjust the abnormal operation offset parameter in the liquid level detection result based on the response feedback consistency mechanism, map the liquid level detection signal to the liquid handling operation-response state chain to correct the contribution coefficient of the equipment status monitoring weight, and update the equipment status monitoring scheme.

[0007] Specifically, the method for constructing the liquid surface pipetting signal window is as follows: In the pipetting operation detection parameters, candidate liquid surface response points in the pipetting stroke domain are identified by a time-series change point detection algorithm. Multimodal signal features are extracted with the candidate liquid surface response points as the center, and a multimodal detection attenuation factor is introduced to cover the pre-response stage before liquid surface contact, the liquid surface crossing transient, and the recovery stage after liquid surface stabilization, thereby obtaining the liquid surface pipetting signal window.

[0008] Specifically, the multimodal detection attenuation factor is a piecewise linear attenuation function with the time difference between the detection modal response time and the current pipetting operation evaluation time and the operation stroke offset as independent variables. For each detection modality, the time difference and operation stroke offset are substituted into the corresponding piecewise linear attenuation function as independent variables, and the response characteristics of the detection modality are weighted to obtain the response correlation value under the current pipetting operation.

[0009] Specifically, the method for generating the pipetting operation feature sequence is as follows: the response points are dynamically weighted and fused based on the multimodal detection attenuation factor, and the fused response points are organized according to the pipetting operation time sequence to form a time-series response lag signal. According to the preset operation stage division rules, the liquid surface response in each stage is mapped to the corresponding operation stage node to extract the discrete response vector caused by the operation disturbance, thereby obtaining the pipetting operation feature sequence.

[0010] Specifically, the method for constructing the pipetting response sensing matrix is ​​as follows: based on the pipetting operation feature sequence, the feature sequence is divided into stages according to the pipetting operation stroke and time dimension, and the different pipetting operation stages and corresponding multimodal response features are used as the matrix dimension to construct the pipetting response sensing matrix of the stage correlation and temporal evolution relationship of the liquid surface response features in the pipetting process.

[0011] Specifically, the pipetting response sensing matrix is ​​layered according to the pipetting operation stage, and the response consistency index and stability index under different pipetting operation stages are detected. The response consistency index is a statistical measure of the liquid surface response of different detection modes in the same pipetting operation stage in terms of temporal position, response amplitude and change trend. The stability index is a measure of the fluctuation amplitude of the liquid surface response of the detection mode in continuous pipetting operation or adjacent operation stages. The response correlation value is used as the adjustment contribution benchmark, and the consistency index and stability index are jointly weighted and fused to obtain the equipment status monitoring weight.

[0012] Specifically, the method for modeling the liquid surface detection process is as follows: the liquid handling operation feature sequence is used as the operation behavior input, the liquid handling response sensing matrix is ​​used as the liquid surface response structure, and the equipment status monitoring weight is introduced to constrain the response contribution caused by equipment status fluctuations. The multimodal liquid surface response is segmented and modeled according to the liquid handling operation stage. Within each liquid handling operation stage, the liquid surface detection signal is truncated within the stage, with the time interval of the corresponding stage in the liquid handling operation feature sequence as a constraint, to obtain the instantaneous amplitude characteristics and temporal distribution density characteristics of the liquid surface detection signal within that stage. The temporal distribution density characteristics are then parametrically modeled to extract the discreteness and tail distribution characteristics of the liquid surface detection signal within that liquid handling operation stage, thereby constructing a liquid surface detection response model.

[0013] Specifically, the quantile statistical evaluation algorithm, based on the liquid surface detection response model during the liquid surface detection process, performs joint statistics on the probability distribution of the liquid surface detection signal in the corresponding liquid transfer operation stage, and calculates the quantile of the liquid surface detection signal in the liquid transfer operation stage according to the preset quantile ratio, thereby quantifying the degree of deviation of the liquid surface detection signal from the normal liquid surface response range of the response model within the corresponding stage and from the abnormal fluctuation range deviating from the response model.

[0014] Specifically, the output process of the liquid level detection result is as follows: the equipment status monitoring weight is used as a status sensitive adjustment factor to jointly evaluate the instantaneous amplitude characteristics and temporal distribution characteristics of the liquid level detection signal in the current liquid transfer operation stage, find the quantile statistical benchmark interval in the corresponding operation state, and remove the detection response points that fall into the abnormal quantile interval according to the quantile statistical benchmark interval, and output the liquid level detection result of the liquid level response in the candidate interval in the time dimension and travel dimension.

[0015] Specifically, the response feedback consistency mechanism includes liquid level response mapping and feedback correction adjustment; The liquid surface response mapping maps the state-sensitive corrected liquid surface detection signal to the operation behavior characteristics of the corresponding liquid handling operation stage, calculates the deviation index between the liquid surface response and the expected liquid surface response mode in each operation stage, and obtains the liquid handling operation-liquid surface response mapping chain. The feedback correction adjustment, based on the weighted correction result, maps the deviation index of the liquid surface response inversely to the abnormal operation offset parameter space. It dynamically corrects the liquid surface response that deviates too much or does not meet the continuity constraint, and recalculates the contribution of the liquid surface response in combination with the corrected liquid surface detection result, and updates it to the equipment condition monitoring scheme and the liquid surface detection signal mapping chain.

[0016] Specifically, the method for correcting the contribution coefficient of the pipetting operation-response state chain is as follows: the liquid surface response signal in each pipetting operation stage is mapped to the corresponding operation stage node, the liquid surface response deviation index under that node is calculated, the original contribution coefficient and the corresponding equipment status monitoring weight are jointly initialized, for chain nodes with large deviation, the contribution coefficient of low stability mode is reduced by weighted attenuation function, and the corrected contribution coefficient is applied to the liquid surface response mapping chain to generate a new pipetting operation-response state chain.

[0017] Specifically, a multimodal liquid level detection system for pipetting operations includes: The pipetting operation feature sequence construction module acquires pipetting operation detection parameters, identifies candidate liquid surface response points through a time-series change point detection algorithm, generates a liquid surface detection signal, constructs a liquid surface pipetting signal window based on the liquid surface detection signal, introduces a multi-modal detection attenuation factor to calculate the response correlation value corresponding to the current pipetting operation, and obtains the pipetting operation feature sequence. Equipment status monitoring scheme generation module: Constructs a pipetting response sensing matrix, obtains the temporal evolution relationship of the pipetting operation feature sequence under different pipetting stages, and queries the equipment status monitoring weight corresponding to the current pipetting stage based on the response correlation value to generate an equipment status monitoring scheme; The detection modeling and sensitivity correction module executes the equipment status monitoring scheme, models the liquid level detection process, calculates the quantiles of the liquid level detection device under different operating conditions through the quantile statistical evaluation algorithm, and performs status sensitivity correction on the normal liquid level response and abnormal response caused by equipment status fluctuations in the liquid level detection signal based on the quantiles, and outputs the liquid level detection results. Response feedback consistency correction module: Based on the response feedback consistency mechanism, adjust the abnormal operation offset parameter in the liquid level detection result, map the liquid level detection signal to the liquid handling operation-response state chain to correct the contribution coefficient of the equipment status monitoring weight, and update the equipment status monitoring scheme.

[0018] The beneficial effects of this invention are as follows: This invention achieves high-precision, phase-aware dynamic monitoring of liquid level changes by constructing a closed-loop analysis system of multimodal liquid level detection and liquid handling operation feature sequences. Compared with existing technologies that rely solely on single-modal signals or fixed thresholds for judgment, this invention can integrate multimodal signals such as optical, capacitive, and pressure signals, and dynamically weight response points using multimodal detection attenuation factors to improve the capture rate of key liquid level responses, while effectively suppressing the interference of noise and transient disturbances on the detection results.

[0019] By employing a phased pipetting operation-response modeling and quantile statistical evaluation algorithm, this invention can quantify the quantiles of the liquid surface response and perform state-sensitive correction for abnormal responses, thereby improving the reliability and stability of the liquid surface detection results. Simultaneously, by mapping the liquid surface detection results to the pipetting operation-response state chain and performing closed-loop adaptive correction on the contribution coefficients, a high degree of consistency between the liquid surface response signal and the operational behavior is achieved, improving the system's adaptability under different operating conditions and equipment states.

[0020] Furthermore, the liquid level detection results generated by this invention possess multi-dimensional information such as stage identifiers, response confidence intervals, and modal contribution weights, providing usable and refined input data for pipetting control, experimental data feedback, and equipment status monitoring. The overall method has technical advantages such as traceable operation stages, modal weighted adaptive processing, dynamic correction of abnormal responses, and closed-loop feedback optimization, which can significantly improve the accuracy, stability, and reliability of automated liquid handling in laboratories. Attached Figure Description

[0021] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0022] Figure 1 This is a schematic diagram of the framework of a multimodal liquid level detection method and system for pipetting operations according to the present invention.

[0023] Figure 2 This is a schematic diagram of a multimodal liquid level detection method and response feedback consistency mechanism in a system for pipetting operations according to the present invention. Detailed Implementation

[0024] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0025] Please see Figure 1 A multimodal liquid level detection method for pipetting operations: S1: Obtain the detection parameters of the pipetting operation, identify candidate liquid surface response points through the time-series change point detection algorithm, generate liquid surface detection signals, construct a liquid surface pipetting signal window based on the liquid surface detection signals, introduce a multi-modal detection attenuation factor to calculate the response correlation value corresponding to the current pipetting operation, and obtain the pipetting operation feature sequence; S2: Construct a pipetting response sensing matrix, obtain the temporal evolution relationship of the pipetting operation feature sequence under different pipetting stages, and query the equipment status monitoring weight corresponding to the current pipetting stage based on the response correlation value to generate an equipment status monitoring scheme; S3: Execute the equipment status monitoring scheme, model the liquid level detection process, calculate the quantiles of the liquid level detection device under different operating states through the quantile statistical evaluation algorithm, and perform state-sensitive correction on the normal liquid level response and abnormal response caused by equipment status fluctuations in the liquid level detection signal based on the quantiles, and output the liquid level detection results. S4: Adjust the abnormal operation offset parameter in the liquid level detection result based on the response feedback consistency mechanism, map the liquid level detection signal to the liquid handling operation-response state chain to correct the contribution coefficient of the equipment status monitoring weight, and update the equipment status monitoring scheme.

[0026] In this embodiment, the method for constructing the liquid surface pipetting signal window is as follows: in the pipetting operation detection parameters, candidate liquid surface response points in the pipetting stroke domain are identified by a time-series change point detection algorithm, multimodal signal features are extracted with the candidate liquid surface response points as the center, and a multimodal detection attenuation factor is introduced to cover the pre-response stage before liquid surface contact, the liquid surface crossing transient, and the recovery stage after liquid surface stabilization, thereby obtaining the liquid surface pipetting signal window.

[0027] In this embodiment, the multimodal detection attenuation factor is a piecewise linear attenuation function with the time difference between the detection modal response time and the current pipetting operation evaluation time and the operation stroke offset as independent variables. For each detection modality, the time difference and operation stroke offset are substituted into the corresponding piecewise linear attenuation function as independent variables, and the response characteristics of the detection modality are weighted to obtain the response correlation value under the current pipetting operation.

[0028] In this embodiment, an automated pipetting workstation (such as the pipette equipped with a Hamilton STAR series robotic arm) is used as an example for high-throughput sample processing, such as aspirating cell culture medium from a 96-well plate. Pipetting operations involve precise detection of the liquid level to avoid air intake or liquid spillage. This implementation is based on the aforementioned multimodal liquid level detection method, employing a capacitance sensor (monitoring capacitance changes), a pressure sensor (monitoring aspiration pressure fluctuations), and an optical sensor (monitoring light reflection intensity) as multimodal detection sources. The pipetting operation detection parameters include real-time acquired capacitance values ​​(pF), pressure values ​​(Pa), and optical intensity values ​​(lux). The entire process is performed during the pipetting stroke (from the pipette tip entering the container to liquid contact and aspiration), assuming a pipetting volume of 200 μL, a standard 1.5 mL centrifuge tube, and a liquid level of approximately 10 mm.

[0029] Acquire pipetting operation detection parameters and generate pipetting operation feature sequences Acquiring detection parameters: As the pipette begins to descend into the centrifuge tube, multimodal data is acquired in real time. The capacitance sensor acquires the increase in capacitance as the pipette tip approaches the liquid surface (from an initial 5 pF to a peak of 15 pF), the pressure sensor acquires the pressure decrease during liquid aspiration (from atmospheric pressure 101325 Pa to negative pressure 98000 Pa), and the optical sensor acquires the change in reflected light intensity from the liquid surface (from a background of 50 lux to a peak of 200 lux). The acquisition frequency is 100 Hz, generating time-series data sequences: capacitance sequence C(t), pressure sequence P(t), and optical sequence O(t), where t is a timestamp (unit: ms).

[0030] Identifying candidate liquid level response points: Apply a time-series change point detection algorithm (such as Pelletier change point detection, using a threshold of 0.5 standard deviation multiples). In the pipetting stroke domain (tip movement distance 0-15mm), identify abrupt change points in the capacitance sequence (t=150ms, capacitance jump ΔC=8pF), negative pressure initiation points in the pressure sequence (t=152ms, ΔP=-5000Pa), and intensity peak points in the optical sequence (t=151ms, ΔO=120lux) as candidate liquid level response points.

[0031] Generate liquid surface detection signal: merge candidate points to generate comprehensive liquid surface detection signal L(t) = w1C(t) + w2P(t) + w3*O(t), with initial weights w1=w2=w3=1 / 3.

[0032] A liquid level transfer signal window was constructed: Multimodal signal features were extracted with a window width of ±50 ms, centered on the candidate liquid level response point (average t = 151 ms). The window covered the pre-response stage (tip approaching the liquid level, t = 100-150 ms, characteristic: gradual increase trend), the liquid level crossing transient (t = 150-152 ms, characteristic: abrupt peak change), and the recovery stage (t = 152-200 ms, characteristic: stable decay). A multimodal detection attenuation factor α_m (piecewise linear attenuation function: if time difference Δt < 10 ms, then α = 1; 10 ms ≤ Δt < 30 ms, then α = 1 - 0.02Δt; Δt ≥ 30 ms, then α = 0.5) was introduced, while also considering the operating stroke offset Δd (unit: mm; if Δd < 1 mm, then no attenuation; otherwise, attenuation of 0.1Δd). For each mode, calculate the response correlation value R_m = α_m * response characteristic (e.g., for the capacitive mode, R_c = 0.9 * 10pF = 9).

[0033] The pipetting operation feature sequence is obtained: a dynamically weighted fusion response (fusion response point F(t) = Σ R_m), which is organized into a time-series response lag signal (considering a 5ms lag). According to the preset operation stage division rules (pre-response, crossover, recovery), the discrete response vector of each stage is mapped (e.g., pre-response vector: [increment rate 0.1pF / ms, fluctuation variance 0.2]), forming the pipetting operation feature sequence S = [vector1, vector2, vector3].

[0034] Construct a pipetting response sensing matrix and generate an equipment condition monitoring scheme. Constructing a pipetting response sensing matrix: Based on the feature sequence S, stages are divided according to pipetting distance (0-5mm pre-response, 5-10mm crossing, 10-15mm recovery) and time dimension (0-100ms, 100-200ms, 200-300ms). A matrix M (3x9, e.g., M[1,1]=t=151ms position, M[1,2]=peak 12pF amplitude) is constructed using the stage (rows) and multimodal response features (columns: temporal position, amplitude, trend) as dimensions. The matrix obtains the temporal evolution relationship, such as the trend continuity from pre-response to crossing (correlation coefficient 0.85).

[0035] Query the equipment status monitoring weights: Calculate the response consistency index (statistical mean of temporal position deviation <2ms, amplitude deviation <10%, and trend correlation >0.8 within the same stage, 0.75) and the stability index (measurement of fluctuation amplitude <5% during continuous operation, 0.9) based on the stage-level hierarchical matrix M. Using the response correlation value R (average 8.5) as the adjustment benchmark, perform joint weighted fusion (weight = 0.6 consistency + 0.4 stability * R / 10) to obtain the monitoring weight W = [capacitance 0.4, pressure 0.35, optics 0.25].

[0036] Generate an equipment status monitoring scheme: The scheme includes a weight W applied to subsequent monitoring, and threshold settings (such as an abnormal fluctuation threshold of ±15%).

[0037] Implement the equipment condition monitoring plan and output the liquid level detection results. Modeling the liquid surface detection process: Using the feature sequence S as input, matrix M as the response structure, and weights W to constrain modal contributions. Segmented modeling: During the pre-response stage, L (t=0-100ms) is extracted to obtain instantaneous amplitude features (mean 8pF) and temporal distribution density features (Gaussian distribution μ=50ms, σ=10ms). Parametric modeling extracts the degree of dispersion (entropy 1.2) and tail distribution characteristics (skewness 0.3), calculates the temporal concentration (0.85), and constructs a response model (such as a Gaussian mixture model GMM).

[0038] Quantile statistical evaluation: Based on the model, the probability distribution of L(t) within each stage is statistically analyzed. Preset quantile ratios (0.25, 0.5, 0.75), calculate quantiles (Q0.25=5pF, Q0.5=10pF, Q0.75=14pF), and quantify the normal response range (5-14pF) and abnormal offsets (<5pF or >14pF).

[0039] State-sensitive correction: Using weight W as an adjustment factor, the instantaneous amplitude and distribution characteristics are evaluated. The quantile reference interval (Q0.25-Q0.75) is identified, and outliers are eliminated (e.g., a -6000Pa offset at t=160ms in the pressure mode is considered to be caused by equipment vibration). Output liquid level detection results: Liquid level position at time t=151ms, stroke 9.5mm.

[0040] Adjust and update the scheme based on the response feedback consistency mechanism. Liquid surface response mapping: The corrected signal L(t) is mapped to the operational behavior characteristics (downward movement speed 2 mm / s, aspiration rate 50 μL / s), and the deviation index (the deviation between the actual response and the expected mode, such as capacitance deviation of 5%) is calculated. The pipetting operation-liquid surface response mapping chain is obtained (chain nodes: pre-response deviation 3%, crossing deviation 2%, recovery deviation 4%).

[0041] Feedback correction adjustment: Reverse mapping deviation to the abnormal operation offset parameter space (offset parameters: time lag ±3ms, amplitude ±10%). Dynamically correct large deviation nodes (recovery stage deviation 4% > threshold 3%, corrected to 3%). Recalculate modal contribution (capacitance increased to 0.45), update monitoring scheme (new weight W'=[0.45, 0.3, 0.25]) and mapping chain.

[0042] Pipette operation-response state chain correction contribution coefficient: Map each stage signal to a node and calculate the deviation (pre-response node deviation 3%). Jointly initialize the original contribution coefficients (initial [0.33, 0.33, 0.33]) with weight W, and weighted attenuate the low-stability mode (optical stability 0.8 < 0.9, attenuation function β = 1 - 0.1 * (deviation / threshold) = 0.9). The correction coefficient is applied to the chain to generate a new chain (overall deviation reduced to 2.5%).

[0043] In this embodiment, the method for generating the pipetting operation feature sequence is as follows: the response points are dynamically weighted and fused based on the multimodal detection attenuation factor, and the fused response points are organized according to the pipetting operation time sequence to form a time-series response lag signal. According to the preset operation stage division rules, the liquid surface response in each stage is mapped to the corresponding operation stage node to extract the discrete response vector caused by the operation disturbance, thereby obtaining the pipetting operation feature sequence.

[0044] In this embodiment, in a laboratory automated pipetting system, liquid level operation data is acquired through a multimodal sensor, including: 1. Acquisition of detection parameters for pipetting operation Optical liquid level height signal h opt (t); Capacitor liquid level signal h cap (t); Pressure sensor signal p(t); The displacement of the pipette piston Δx(t); The time series of pipetting operations is t∈[0,T].

[0045] 2. Algorithm for detecting time-series change points For liquid level signal s m (t) Extract key response points using a segmented temporal change point detection algorithm: , Where w is the length of the sliding window. For the detected change point time, K is the total number of change points. The change points are used as key nodes to construct the liquid level transfer signal window, ensuring that the operation feature sequence contains only high-information signal points.

[0046] 3. Multimodal detection attenuation factor and dynamic weighted fusion For each modal response, a piecewise linear decay function is introduced: , Where Δt=t k - β represents the time difference between the current sampling point and the point of change. m The modal attenuation coefficient is adaptively adjusted by the system based on modal stability and historical response. The fused signal is organized according to the pipetting operation time series and a preliminary pipetting operation feature sequence is generated based on the operation stage labeling.

[0047] 4. Adjusting the contribution baseline and constructing the pipetting operation-response state chain To dynamically adjust the contribution coefficients, an adjustment contribution baseline matrix W is defined. m Based on the liquid surface response deviation δ m (t k Calculate the correction factor and map the liquid level response at each stage to the operation-response state chain: , , in, λ is the initial contribution coefficient, λ is the adjustment gain, and δ is the initial contribution coefficient. m (t k )=∣s fused (t k )−s ref (t k )∣,s ref (t k The state chain is used as the stage reference response signal, and is used for subsequent modal contribution adaptive correction and abnormal response correction.

[0048] 5. Quantile Statistical Evaluation Algorithm and Quantile Statistical Benchmark Interval Within each operational phase ϕ=p, the quantile Q of the liquid surface response signal is statistically analyzed. γ An abnormal offset response is identified by setting a quantile statistical benchmark interval. The abnormal response is weighted and suppressed according to the state-sensitive correction mechanism, with high consistency mode enhancement and low consistency mode suppression, thus completing the closed-loop optimization.

[0049] In this embodiment, the method for constructing the pipetting response sensing matrix is ​​as follows: based on the pipetting operation feature sequence, the feature sequence is divided into stages according to the pipetting operation stroke and time dimension, and the different pipetting operation stages and corresponding multimodal response features are used as the matrix dimension to construct the pipetting response sensing matrix of the stage correlation and temporal evolution relationship of the liquid surface response features in the pipetting process.

[0050] In this embodiment, the pipetting response sensing matrix is ​​layered according to the pipetting operation stage, and the response consistency index and stability index under different pipetting operation stages are detected. The response consistency index is a statistical measure of the liquid surface response of different detection modes in the same pipetting operation stage in terms of temporal position, response amplitude and change trend. The stability index is a measure of the fluctuation amplitude of the liquid surface response of the detection mode in continuous pipetting operation or adjacent operation stages. The response correlation value is used as the adjustment contribution benchmark, and the consistency index and stability index are jointly weighted and fused to obtain the equipment status monitoring weight.

[0051] In this embodiment, the method for modeling the liquid surface detection process is as follows: the liquid handling operation feature sequence is used as the operation behavior input, the liquid handling response sensing matrix is ​​used as the liquid surface response structure, and the equipment status monitoring weight is introduced to constrain the response contribution caused by equipment status fluctuations. The multimodal liquid surface response is segmented and modeled according to the liquid handling operation stage. Within each liquid handling operation stage, the liquid surface detection signal is truncated within the stage, with the time interval of the corresponding stage in the liquid handling operation feature sequence as a constraint, to obtain the instantaneous amplitude characteristics and temporal distribution density characteristics of the liquid surface detection signal within the stage. The temporal distribution density characteristics are then parametrically modeled to extract the discreteness and tail distribution characteristics of the liquid surface detection signal within the liquid handling operation stage, thereby constructing a liquid surface detection response model.

[0052] In this embodiment, the quantile statistical evaluation algorithm performs joint statistics on the probability distribution of the liquid surface detection signal in the corresponding liquid transfer operation stage based on the liquid surface detection response model during the liquid surface detection process, and calculates the quantile of the liquid surface detection signal in the liquid transfer operation stage according to the preset quantile ratio, thereby quantifying the degree of deviation of the liquid surface detection signal from the normal liquid surface response range of the response model and the abnormal fluctuation range that deviates from the response model within the corresponding stage.

[0053] In this embodiment, a liquid level detection response model is constructed. This model integrates multimodal signals, characteristics of the liquid handling stage, and equipment state weights. The dynamic representation of the liquid level response is achieved through segmented statistical modeling, time-series feature extraction, and closed-loop correction.

[0054] 1. Stage Feature Modeling Layer This layer is used to capture the amplitude and time-series distribution characteristics of the liquid surface response during the pipetting operation phase, for multimodal liquid surface signals s m (t k The signals are fused together to generate continuous time-series signals within the stage. The liquid level signal at each stage is truncated into a subsequence of ϕ=p, and the instantaneous amplitude feature A is extracted. p (t k ) and temporal distribution density characteristics D p (τ), construct the model: , in, , σ is the average value over the period. A,p ,σ D,p , where Z is the standard deviation. p These are the normalization coefficients. This layer is primarily used to quantify the statistical characteristics of the liquid surface response in both amplitude and time dimensions, providing a foundation for identifying anomalous responses.

[0055] 2. Operation-Response State Chain Modeling Layer This layer is used to dynamically couple and model the multi-stage liquid surface response and pipetting operation behavior, and realize closed-loop adaptive optimization of modal contribution. It constructs a pipetting operation-response state chain, calculates the liquid surface response deviation for each stage node in the chain, corrects the contribution coefficient based on the deviation and the initial contribution coefficient using a weighted exponential adjustment formula, and applies stage continuity constraints to the corrected contribution coefficient. It outputs the closed-loop optimized operation-response state chain, providing input for liquid surface anomaly correction, quantile statistical evaluation, and equipment condition monitoring weight update.

[0056] In this embodiment, the output process of the liquid level detection result is as follows: the equipment status monitoring weight is used as a status sensitive adjustment factor to jointly evaluate the instantaneous amplitude characteristics and temporal distribution characteristics of the liquid level detection signal in the current liquid transfer operation stage, find the quantile statistical benchmark interval in the corresponding operation state, and remove the detection response points that fall into the abnormal quantile interval according to the quantile statistical benchmark interval, and output the liquid level detection result of the liquid level response in the candidate interval in the time dimension and travel dimension.

[0057] In this embodiment, the response feedback consistency mechanism includes liquid surface response mapping and feedback correction adjustment; The liquid surface response mapping maps the state-sensitive corrected liquid surface detection signal to the operation behavior characteristics of the corresponding liquid handling operation stage, calculates the deviation index between the liquid surface response and the expected liquid surface response mode in each operation stage, and obtains the liquid handling operation-liquid surface response mapping chain. The feedback correction adjustment, based on the weighted correction result, maps the deviation index of the liquid surface response inversely to the abnormal operation offset parameter space, dynamically corrects the liquid surface response that deviates too much or does not meet the continuity constraint, and recalculates the contribution of the liquid surface response in combination with the corrected liquid surface detection result, and updates it to the equipment condition monitoring scheme and the liquid surface detection signal mapping chain.

[0058] In this embodiment, the method for correcting the contribution coefficient of the pipetting operation-response state chain is as follows: the liquid surface response signal in each pipetting operation stage is mapped to the corresponding operation stage node, the liquid surface response deviation index under the node is calculated, the original contribution coefficient and the corresponding equipment status monitoring weight are jointly initialized, for chain nodes with large deviation, the contribution coefficient of the low stability mode is reduced by the weighted attenuation function, and the corrected contribution coefficient is applied to the liquid surface response mapping chain to generate a new pipetting operation-response state chain.

[0059] This invention also provides a multimodal liquid level detection system for pipetting operations, specifically including: The pipetting operation feature sequence construction module acquires pipetting operation detection parameters, identifies candidate liquid surface response points through a time-series change point detection algorithm, generates a liquid surface detection signal, constructs a liquid surface pipetting signal window based on the liquid surface detection signal, introduces a multi-modal detection attenuation factor to calculate the response correlation value corresponding to the current pipetting operation, and obtains the pipetting operation feature sequence. Equipment status monitoring scheme generation module: Constructs a pipetting response sensing matrix, obtains the temporal evolution relationship of the pipetting operation feature sequence under different pipetting stages, and queries the equipment status monitoring weight corresponding to the current pipetting stage based on the response correlation value to generate an equipment status monitoring scheme; The detection modeling and sensitivity correction module executes the equipment status monitoring scheme, models the liquid level detection process, calculates the quantiles of the liquid level detection device under different operating conditions through the quantile statistical evaluation algorithm, and performs status sensitivity correction on the normal liquid level response and abnormal response caused by equipment status fluctuations in the liquid level detection signal based on the quantiles, and outputs the liquid level detection results. Response feedback consistency correction module: Based on the response feedback consistency mechanism, adjust the abnormal operation offset parameter in the liquid level detection result, map the liquid level detection signal to the liquid handling operation-response state chain to correct the contribution coefficient of the equipment status monitoring weight, and update the equipment status monitoring scheme.

[0060] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A multimodal liquid level detection method for pipetting operations, characterized in that, include: S1: Obtain the detection parameters of the pipetting operation, identify candidate liquid surface response points through the time-series change point detection algorithm, generate liquid surface detection signals, construct a liquid surface pipetting signal window based on the liquid surface detection signals, introduce a multi-modal detection attenuation factor to calculate the response correlation value corresponding to the current pipetting operation, and obtain the pipetting operation feature sequence; S2: Construct a pipetting response sensing matrix, obtain the temporal evolution relationship of the pipetting operation feature sequence under different pipetting stages, and query the equipment status monitoring weight corresponding to the current pipetting stage based on the response correlation value to generate an equipment status monitoring scheme; S3: Execute the equipment status monitoring scheme, model the liquid level detection process, calculate the quantiles of the liquid level detection device under different operating states through the quantile statistical evaluation algorithm, and perform state-sensitive correction on the normal liquid level response and abnormal response caused by equipment status fluctuations in the liquid level detection signal based on the quantiles, and output the liquid level detection results. S4: Adjust the abnormal operation offset parameter in the liquid level detection result based on the response feedback consistency mechanism, map the liquid level detection signal to the liquid handling operation-response state chain to correct the contribution coefficient of the equipment status monitoring weight, and update the equipment status monitoring scheme.

2. The method according to claim 1, characterized in that, The method for constructing the liquid surface pipetting signal window is as follows: In the pipetting operation detection parameters, candidate liquid surface response points in the pipetting stroke domain are identified by a time-series change point detection algorithm. Multimodal signal features are extracted with the candidate liquid surface response points as the center, and a multimodal detection attenuation factor is introduced to cover the pre-response stage before liquid surface contact, the liquid surface crossing transient, and the recovery stage after liquid surface stabilization, thereby obtaining the liquid surface pipetting signal window.

3. The method according to claim 1, characterized in that, The multimodal detection attenuation factor is a piecewise linear attenuation function with the time difference between the detection modal response time and the current pipetting operation evaluation time and the operation stroke offset as independent variables. For each detection modality, the time difference and operation stroke offset are substituted into the corresponding piecewise linear attenuation function as independent variables, and the response characteristics of the detection modality are weighted to obtain the response correlation value under the current pipetting operation.

4. The method according to claim 1, characterized in that, The method for generating the pipetting operation feature sequence is as follows: the response points are dynamically weighted and fused based on the multimodal detection attenuation factor, and the fused response points are organized according to the pipetting operation time sequence to form a time-series response lag signal. According to the preset operation stage division rules, the liquid surface response in each stage is mapped to the corresponding operation stage node to extract the discrete response vector caused by the operation disturbance, thereby obtaining the pipetting operation feature sequence.

5. The method according to claim 2, characterized in that, The method for constructing the pipetting response sensing matrix is ​​as follows: based on the pipetting operation feature sequence, the feature sequence is divided into stages according to the pipetting operation stroke and time dimension, and the different pipetting operation stages and corresponding multimodal response features are used as the matrix dimension to construct the pipetting response sensing matrix of the stage correlation and temporal evolution relationship of the liquid surface response features in the pipetting process.

6. The method according to claim 5, characterized in that, The calculation method for the equipment status monitoring weight is as follows: the pipetting response sensing matrix is ​​layered according to the pipetting operation stage, and the response consistency index and stability index under different pipetting operation stages are detected. The response consistency index is a statistical measure of the liquid surface response of different detection modes in the same pipetting operation stage in terms of temporal position, response amplitude and change trend. The stability index is a measure of the fluctuation amplitude of the liquid surface response of the detection mode in continuous pipetting operation or adjacent operation stages. The response correlation value is then used as the adjustment contribution benchmark, and the consistency index and stability index are jointly weighted and fused to obtain the equipment status monitoring weight.

7. The method according to claim 4, characterized in that, The method for modeling the liquid surface detection process is as follows: the liquid handling operation feature sequence is used as the operation behavior input, the liquid handling response perception matrix is ​​used as the liquid surface response structure, and the equipment status monitoring weight is introduced to constrain the response contribution caused by equipment status fluctuations. The multimodal liquid surface response is segmented and modeled according to the liquid handling operation stage. Within each pipetting operation stage, the liquid surface detection signal is truncated within the stage, constrained by the time interval of the corresponding stage in the pipetting operation feature sequence. The instantaneous amplitude characteristics and temporal distribution density characteristics of the liquid surface detection signal within the stage are obtained. The temporal distribution density characteristics are then parametrically modeled to extract the discreteness and tail distribution characteristics of the liquid surface detection signal within the pipetting operation stage, thereby constructing a liquid surface detection response model.

8. The method according to claim 2, characterized in that, During the liquid surface detection process, the quantile statistical evaluation algorithm, based on the liquid surface detection response model, performs joint statistics on the probability distribution of the liquid surface detection signal in the corresponding liquid transfer operation stage, and calculates the quantile of the liquid surface detection signal in the liquid transfer operation stage according to the preset quantile ratio, thereby quantifying the degree of deviation of the liquid surface detection signal from the normal liquid surface response range of the response model within the corresponding stage and from the abnormal fluctuation range of the response model.

9. The method according to claim 4, characterized in that, The output process of the liquid level detection results is as follows: the equipment status monitoring weight is used as a status-sensitive adjustment factor to jointly evaluate the instantaneous amplitude characteristics and temporal distribution characteristics of the liquid level detection signal in the current liquid transfer operation stage, find the quantile statistical benchmark interval in the corresponding operation state, and remove the detection response points that fall into the abnormal quantile interval according to the quantile statistical benchmark interval, and output the liquid level detection results of the liquid level response in the candidate interval in the time dimension and travel dimension.

10. The method according to claim 1, characterized in that, The response feedback consistency mechanism includes liquid level response mapping and feedback correction adjustment; The liquid surface response mapping maps the state-sensitive corrected liquid surface detection signal to the operation behavior characteristics of the corresponding liquid handling operation stage, calculates the deviation index between the liquid surface response and the expected liquid surface response mode in each operation stage, and obtains the liquid handling operation-liquid surface response mapping chain. The feedback correction adjustment maps the deviation index of the liquid surface response to the abnormal operation offset parameter space based on the weighted correction result of the response point. It dynamically corrects the liquid surface response that deviates too much or does not meet the continuity constraint, and recalculates the contribution of the liquid surface response in combination with the corrected liquid surface detection result, and updates it to the equipment condition monitoring scheme and the liquid surface detection signal mapping chain.

11. The method according to claim 1, characterized in that, The method for correcting the contribution coefficient of the pipetting operation-response state chain is as follows: the liquid surface response signal in each pipetting operation stage is mapped to the corresponding operation stage node, the liquid surface response deviation index under the node is calculated, the original contribution coefficient and the corresponding equipment status monitoring weight are jointly initialized, for chain nodes with large deviation, the contribution coefficient of low stability mode is reduced by weighted attenuation function, and the corrected contribution coefficient is applied to the liquid surface response mapping chain to generate a new pipetting operation-response state chain.

12. A multimodal liquid level detection system for pipetting operations, used to perform the method as described in any one of claims 1-11, characterized in that, include: The pipetting operation feature sequence construction module acquires pipetting operation detection parameters, identifies candidate liquid surface response points through a time-series change point detection algorithm, generates a liquid surface detection signal, constructs a liquid surface pipetting signal window based on the liquid surface detection signal, introduces a multi-modal detection attenuation factor to calculate the response correlation value corresponding to the current pipetting operation, and obtains the pipetting operation feature sequence. Equipment status monitoring scheme generation module: Constructs a pipetting response sensing matrix, obtains the temporal evolution relationship of the pipetting operation feature sequence under different pipetting stages, and queries the equipment status monitoring weight corresponding to the current pipetting stage based on the response correlation value to generate an equipment status monitoring scheme; The detection modeling and sensitivity correction module executes the equipment status monitoring scheme, models the liquid level detection process, calculates the quantiles of the liquid level detection device under different operating conditions through the quantile statistical evaluation algorithm, and performs status sensitivity correction on the normal liquid level response and abnormal response caused by equipment status fluctuations in the liquid level detection signal based on the quantiles, and outputs the liquid level detection results. Response feedback consistency correction module: Based on the response feedback consistency mechanism, adjust the abnormal operation offset parameter in the liquid level detection result, map the liquid level detection signal to the liquid handling operation-response state chain to correct the contribution coefficient of the equipment status monitoring weight, and update the equipment status monitoring scheme.

Images