A method for detecting stratified liquid level

Abstract

The present invention provides a layered liquid level detection method suitable for liquid level identification and height estimation in multi-layer liquid structures. The detection method uses a pressure sensor to collect air pressure changes during the liquid aspiration process, constructs a time-pressure curve, and extracts slope and curvature characteristics through first-order and second-order derivative analysis to identify liquid level mutation points. A jump prediction mechanism based on historical trajectories and physical models is introduced, and combined with Reynolds number correction and bubble interference identification algorithms, intelligent judgment and error tolerance of multi-layer liquid levels are achieved. The system of the present invention has the ability to adaptively adjust parameters and maintain high-precision identification under different liquids, environments, and operating conditions. It is widely applicable to scenarios such as automated experiments, layered sampling, and high-throughput liquid processing.

Description

Technical Field

[0001] The present invention relates to the technical field of stratified liquid level detection, and in particular to a stratified liquid level detection method. Background Art

[0002] During liquid handling operations, accurate identification of the liquid surface is fundamental to achieving functions such as quantitative aspiration, automatic liquid exchange, and layered sampling. This is especially true in complex environments with multiple layers of liquid, such as layered extraction of biological samples, staged sampling of chemical reaction solutions, and quantitative identification of oil-water interfaces.

[0003] Commonly used liquid level detection methods include capacitive, optical, and pressure-surge methods. Capacitive methods are sensitive to conductivity and are easily limited by sample type; optical methods are more sensitive to liquid transparency and external light interference; and pressure-surge methods, while offering advantages such as simple structure and fast response, are susceptible to contamination of non-liquid level signals such as foam interference and pipeline vibration, leading to misidentification or missed identification. Therefore, we propose a layered liquid level detection method. Summary of the Invention

[0004] The purpose of the present invention is to provide a stratified liquid level detection method to address the deficiencies of the prior art and to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A method for detecting a stratified liquid level comprises the following steps:

[0007] S1. Collect the air pressure change data in the pipetting pump, combine the compression feedback characteristics, construct the air pressure compression state composite curve, process and analyze the curve, and extract the air pressure change trend and jump signal characteristics;

[0008] S2. Based on historical data and the time series of the aspiration process, a dynamic threshold prediction model is constructed to generate a time judgment window for liquid level jumps. The real-time liquid level height is calculated using a volume height fitting model optimized with residuals, combining the time dimension of the pipetting process with the aspirated volume and channel structural parameters.

[0009] S3. Fusion of the air pressure change signal and the micro-vibration signal to identify whether there is signal interference caused by bubbles or foam; introduction of the current fluid's Reynolds number, shear rate, and pipeline structure characteristics to construct a flow velocity profile model;

[0010] S4. Establish a liquid surface trajectory history database and train a trajectory prediction model to guide the judgment of liquid surface jumps during the liquid aspiration process; when multiple jump points are detected, identify multiple liquid surface contact positions, infer the corresponding depth of each liquid layer, and complete the layer-by-layer identification of stratified liquids.

[0011] S5. During the liquid aspiration process, if it is determined that the current signal has abnormal fluctuations or insufficient confidence, the reasoning path based on the historical trajectory or fluid physics model is automatically called to correct the liquid level recognition result.

[0012] S6. Automatically adjust the prediction model, judgment threshold and sensing parameters according to changes in the system environmental state to adapt to different liquid properties and detection environments.

[0013] In a preferred embodiment, S1 is specifically:

[0014] Collect the air pressure change data during the liquid aspiration process to form a pressure time series;

[0015] Construct a compression model of the relationship between air pressure and volume to reflect the compression behavior of the air column;

[0016] Calculate the first-order derivative, second-order derivative and curvature value based on the pressure time series to extract the pressure change trend and jump characteristics;

[0017] The extracted features are used to form a standard signal vector, which serves as a candidate basis for identifying liquid surface contact points.

[0018] In a preferred embodiment, S2 is specifically:

[0019] Based on the pressure changes and jump moments during the historical liquid aspiration process, a historical sample database is established;

[0020] The dynamic prediction model is trained based on the samples to determine the time window when the liquid level jump may occur;

[0021] In this judgment window, the suction volume is calculated based on the current suction speed, and the liquid level height is deduced based on the air column structure;

[0022] An error residual optimization function is established, and the credibility of liquid level determination is verified by the deviation between the model height and the volume inversion height.

[0023] In a preferred embodiment, S3 is specifically:

[0024] The air pressure signal and the micro-vibration signal are used to construct a joint feature vector, and the interference recognition model is trained based on the feature vector;

[0025] By identifying the bubble disturbance characteristics, the difference between the bubble desorption signal and the actual liquid level jump signal, the bubble interference can be judged and eliminated;

[0026] The Reynolds number and shear rate are calculated based on the liquid density, channel diameter, flow velocity and viscosity. The liquid motion state is then used to identify whether it is in a laminar or turbulent state, and a segmented identification rule is established.

[0027] In a preferred embodiment, S4 is specifically

[0028] A jump trajectory prediction model is constructed based on historical jump records and aspirated volume, and the next jump moment is predicted in advance through trajectory modeling;

[0029] When multiple jump points are detected, the liquid level depth corresponding to each jump point is calculated by combining the liquid aspiration speed and the instantaneous flow rate to construct a preliminary layered structure;

[0030] Automatically screen or correct layers with abnormal depth spacing, and identify the next jump point by combining jump prediction and volume state to achieve intelligent identification of multi-layer liquid surface structure and dynamic inversion correction of jump points.

[0031] In a preferred embodiment, S5 is specifically:

[0032] Based on the real-time pressure curve characteristics, the abnormal state of the jump is judged and the fault tolerance mechanism is triggered;

[0033] In abnormal conditions, the historical liquid aspiration trajectory is matched first, and the liquid level jump point is predicted using the jump trajectory model under similar conditions. If the historical matching is invalid, a standard jump path is constructed based on the physical model, and the minimum residual value is calculated to correct the jump point.

[0034] Final liquid level recognition and judgment based on multi-source feature fusion;

[0035] Output the recognition confidence level and whether to correct the mark, and use the correction result to optimize the subsequent recognition process.

[0036] In a preferred embodiment, step S6 is specifically as follows:

[0037] Real-time collection of environmental status parameters and operating status data;

[0038] Analyze the sensitivity of the model recognition threshold to each state parameter, and make real-time compensation for key parameters or modify model weights based on the sensitivity results;

[0039] Adaptively adjust the recognition judgment threshold based on the degree of change of state parameters to achieve dynamic update of the threshold as the environment drifts;

[0040] Automatically select the most suitable model structure type based on liquid properties and task requirements;

[0041] When the system environment changes continuously beyond the original applicable range of the model, the model structure migration and parameter recalibration are triggered to perform key feature extraction and reinforcement learning.

[0042] The beneficial effects of the present invention are:

[0043] By constructing a derivative and curvature feature extraction model of the pressure-time curve, combined with jump trajectory modeling and error correction mechanism, it is possible to effectively identify multiple liquid surface contact points and realize accurate judgment of the layered liquid structure, breaking through the limitation of traditional methods that can only identify a single liquid surface; by adopting the Reynolds number and fluid structure feedback mechanism to dynamically correct the judgment threshold, combined with the bubble disturbance feature recognition algorithm, it is possible to effectively eliminate the interference of abnormal signals such as foam and fluctuation on the liquid surface recognition results, and significantly improve the system stability and anti-interference ability; introducing a model parameter adaptive adjustment module, it can dynamically adjust the recognition model structure and threshold according to different liquid types, flow states and environmental parameters, improve the adaptability of the method under complex working conditions, and be suitable for a variety of high-precision liquid processing scenarios such as biology, chemistry, and medicine; when abnormal fluctuations or model deviations occur during the recognition process, the present invention can automatically correct the jump point based on the historical jump trajectory and physical inversion model, realize self-repair of the recognition error, and improve the reliability of the system's continuous and stable operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 This is a schematic diagram of a layered liquid level detection method of the present invention. DETAILED DESCRIPTION

[0045] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0046] Example 1: Figure 1 As shown, this embodiment provides a stratified liquid level detection method, comprising the following steps:

[0047] S1. Signal acquisition and compression model construction: During the aspiration process, the air pressure change data inside the pipette pump is collected, and combined with the compression feedback characteristics of the pipette mechanism, an air pressure-compression state composite curve is constructed. The first-order derivative, second-order derivative, and curvature analysis of this curve are performed to extract the air pressure change trend and jump signal characteristics for preliminary identification of liquid level fluctuations and mutations.

[0048] S2: Dynamic time window prediction and liquid level inversion: Based on historical data and the time series of the aspiration process, a dynamic threshold prediction model is constructed to generate a time judgment window for liquid level jumps; combining the time dimension of the pipetting process with the aspirated volume and the channel structure parameters, the real-time liquid level is inverted and calculated through a volume-height fitting model optimized by residuals.

[0049] S3: Multimodal signal fusion and interference rejection: The air pressure change signal is fused with the micro-vibration signal to identify whether there is signal interference caused by bubbles or foam; at the same time, the Reynolds number, shear rate and pipeline structure characteristics of the current fluid are introduced to construct a flow velocity profile model to further enhance the accuracy and robustness of signal recognition.

[0050] S4. Jump trajectory database construction and intelligent guidance identification: Record the historically identified liquid surface position jump time, establish a liquid surface trajectory history database, and train to form a trajectory prediction model for liquid surface jump judgment and guidance in the subsequent liquid aspiration process; on this basis, when multiple jump points are detected, identify multiple liquid surface contact positions, infer the depth corresponding to each liquid layer, and complete the layer-by-layer identification of the layered liquid structure.

[0051] S5. Abnormal correction and intelligent fault-tolerance mechanism: During the liquid aspiration process, if it is determined that the current signal has abnormal fluctuations or insufficient confidence, the system automatically calls the reasoning path based on the historical trajectory or fluid physics model to correct the liquid level recognition result, thereby improving the overall robustness and fault tolerance of the system.

[0052] S6. Model adaptive optimization and detection environment adaptation: According to changes in the system environmental state, the prediction model, judgment threshold and sensor parameters are automatically adjusted to adapt to different liquid properties and detection environments, thereby improving the universality and application expansion capabilities of the detection system.

[0053] During the aspiration start phase in step S1, the pipette pump is controlled to extract liquid from the target container at a constant flow rate or a specific acceleration. Simultaneously, a pressure sensor located within the pipette channel collects pressure data in real time. The sensor is preferably a highly sensitive micro-differential pressure sensor to ensure that transient jump characteristics can be captured.

[0054] Step S1 includes sub-steps S110-S160 as follows:

[0055] S110, real-time collection of air pressure change data: During the liquid aspiration operation, the air pressure sensor installed inside the pipette pump performs high-frequency sampling on the pressure change of the air column to obtain time series pressure data The sampling frequency is recommended to be set to 100-200 Hz to meet the timing accuracy requirements of dynamic recognition.

[0056] The data collected are: ,in The sampling time interval is usually 5 to 10 ms; is the sampling point index (i.e. the number of points);

[0057] For example, air pressure data is collected every 5 to 10 milliseconds. This sampling frequency balances time resolution with system processing load, ensuring that rapid responses to jump behaviors are accurately captured without causing data redundancy or wasting processing resources.

[0058] S120. Construct a composite model of air pressure compression: During the aspiration process, the air column in the syringe increases in pressure due to volume compression. This relationship conforms to the simplified gas state equation:

[0059]

[0060] in is the volume of the gas column, which is controlled by a stepper motor or piston displacement; is the pressure-volume constant in the initial state, which can be calculated based on the system structure size and control signal , thus forming a pressure-compression state composite curve , as the basis for subsequent identification analysis.

[0061] S130, perform first-order derivative analysis: perform first-order difference processing on the above pressure data to obtain the pressure change rate per unit time (i.e., the first-order derivative):

[0062]

[0063] This indicator reflects the trend of pressure changes in the system and can be used to determine abnormal mutations near the liquid surface contact point. Indicates the The air pressure value at the moment, Indicates the The air pressure value at a moment;

[0064] This indicator represents the degree of change in air pressure per unit time, reflecting the rate of increase or decrease in air pressure during aspiration. This metric is particularly important for liquid surface identification: when the air pressure change suddenly becomes steep, with a large positive or negative slope, it may mean that the tip has contacted or broken through the liquid surface, resulting in a liquid interface jump event. If the slope remains close to zero for a long time, it may be in the bubble stage or static stage. The presence of multiple consecutive slope mutation points can be used for subsequent jump trajectory modeling and layer structure identification.

[0065] S140. Perform second-order derivative and acceleration analysis: Further perform differential processing on the first-order derivative sequence to obtain the acceleration of the pressure change, that is, the second-order derivative:

[0066]

[0067] The second-order derivative can enhance the response sensitivity to short-term sudden changes and help identify the precursory behavior of foam disturbances or liquid surface disturbances.

[0068] Indicates time The first derivative (i.e. slope) of represents the slope at the previous moment, It represents the rate of change of slope (i.e., the acceleration or curvature of air pressure). When the liquid suction system suddenly switches from one fluid to another (such as from air to liquid), not only will the rate of change of air pressure (slope) jump, but the intensity of the jump of the slope (i.e., the second-order derivative) will also increase significantly. The second-order derivative reflects the "change of change" and is more suitable for detecting inflection points, extreme points, or mutation positions in the process of air pressure change.

[0069] S150, calculate the curvature value of the pressure curve: In order to further extract the signal turning trend, calculate the curvature of the pressure curve. The curvature formula is as follows:

[0070]

[0071] When the curvature reaches an extreme value at a certain moment, it indicates that the signal is strongly bent at that point, which usually corresponds to liquid surface contact or bubble interference events.

[0072] When the liquid surface suddenly contacts the detection pipe (for example, from the gas phase into the liquid phase), it will cause the gas pressure to change rapidly; this jump behavior will not only produce a larger slope , which will also cause the slope to change dramatically, that is, Rapid growth;

[0073] curvature Taking the slope and slope change into account can more accurately reflect the intensity of the "bend" point.

[0074] Used alone or In comparison, curvature can suppress the influence of noise through normalization; it responds sensitively to mutation points (such as jump points, interlayer liquid switching points), but has little effect on low-frequency fluctuations (such as bubble disturbances).

[0075] S160, constructing a standardized signal feature vector: To facilitate subsequent recognition and learning processes, the above parameters can be combined into a feature vector:

[0076]

[0077] This vector can be input into the machine learning module as a description of the candidate liquid surface point, or directly used by the rule judgment logic.

[0078] Steps S110-S160 in this embodiment can accurately characterize the compression change characteristics of the gas column during the liquid aspiration process; identify the critical moment of pressure mutation, and provide candidate points for liquid surface recognition; enhance the system's recognition robustness to sudden disturbances, and improve the stability and accuracy of detection; and provide signal foundation support for subsequent sub-modules such as foam recognition and Re correction.

[0079] Step S2 includes sub-steps S210-S250 as follows:

[0080] S210: Build a historical reference database: During long-term operation, the system saves the air pressure curve during the liquid aspiration process and the liquid level recognition moment (i.e., the moment of sudden pressure change) as historical samples to form a time series database:

[0081]

[0082] in, Indicates the The first aspiration The air pressure at a given moment; Indicates the time (or corresponding sampling time index) of the liquid level jump point that has been manually or automatically marked in the data; Indicates the total number of historical samples; Indicates the The pressure time series data of samples, a total of sampling points;

[0083] S220: Constructing a dynamic threshold prediction model: Based on the above database, a sliding window method is used to extract waveform features of multiple local sub-intervals, and a model is trained or rules are constructed to predict the possible occurrence of liquid level jumps within a time window:

[0084]

[0085] For example, based on the current aspiration flow rate and the records of the previous three aspirations, it can be inferred that a sudden change in air pressure may occur between 0.85s and 1.05s. This window range can be used to constrain subsequent recognition processes, improving jump recognition efficiency and anti-interference capabilities.

[0086] S230: Combine the aspiration volume and time dimension to calculate the liquid level

[0087] In the jump window, the system is based on the current pipetting speed. With sampling time , calculate the inhaled volume

[0088]

[0089] Among them, is the instantaneous flow rate of the liquid (known by the system or calculated from the motor parameters), usually in mL / s or L / min; the integral result is the volume of air inhaled before the liquid surface contacts, where Indicates the cumulative volume of liquid aspirated by the system before the trip point is detected during the aspiration process. For discrete moments The flow sampling value on The index corresponding to the time jump point.

[0090] S240: Joint modeling of channel structure and inversion of liquid level

[0091] Assume that the inner diameter of the pipette needle is For a cylindrical channel, the air volume corresponds to the cylindrical space volume:

[0092] , and thus infer:

[0093] Among them, the height It is the vertical distance from the starting position of the needle to the liquid surface, which is the result of liquid surface inversion.

[0094] S250: Error residual optimization mechanism: To reduce the fluctuation error caused by bubbles and vibrations, the system establishes a highly fitted residual function:

[0095]

[0096] And set the residual tolerance threshold , if: , then the liquid level judgment is credible; otherwise, return to the candidate window for re-identification.

[0097] Represents the prediction error, which is used to measure the accuracy of the liquid level recognition model. The unit is usually mm or cm; Indicates the liquid level height predicted by the system based on recognition algorithms (such as jump point recognition, AI model, etc.); It represents the theoretical liquid level calculated based on a known system model (such as a liquid flow model or a liquid suction path model); The smaller the value, the higher the consistency between the identification model and the physical model, and the more stable the system.

[0098] In this embodiment, steps S210-S250 dynamically generate a jump time judgment window based on historical data to improve recognition efficiency and stability; the actual suction volume is inverted by the suction time and flow rate, and the liquid level height is inverted by combining geometric modeling; the model has strong physical interpretability, clear algorithm structure, and strong adaptability; and it can still maintain high recognition robustness in the presence of foam, tilt, changes in liquid viscosity, etc.

[0099] Step S3 can effectively eliminate false triggering caused by bubbles or foam, and improve the accuracy and robustness of liquid level recognition. The method includes the following sub-steps S310-S350:

[0100] S310: Synchronously collect micro-vibration and air pressure change signals

[0101] During the aspiration process, in addition to collecting the real-time pressure change sequence output by the pressure sensor In addition, the system is also connected to a micro accelerometer or MEMS vibration sensor to collect micro vibration signals of the liquid during the needle entering the process. .

[0102] The micro-vibration signal reflects the disturbance changes in the process of fluid entering the needle tube, especially when the bubble bursts or the foam adheres to cause oscillation, it has a strong fluctuation characteristic.

[0103] S320: Constructing multimodal joint feature vectors

[0104] The pressure signal and vibration signal are standardized separately, and the joint feature vector is established synchronously in time:

[0105]

[0106] in : are the first derivative, second derivative and curvature of the pressure curve respectively; Indicates the vibration acceleration at the current moment; It represents the first derivative of the vibration signal and reflects the rate of change of vibration energy.

[0107] S330: Constructing an interference discrimination model:

[0108] By analyzing the characteristic patterns under the interference of foam / bubbles, the system trains or sets the following judgment rules: If the vibration signal is at the point of sudden change in air pressure, Synchronous sudden rise (such as short-term change exceeding the threshold ), which can be determined as foam interference;

[0109] like Extremely large ≈0, it is the real liquid level signal, and support vector machine (SVM), decision tree or lightweight neural network can also be introduced to Perform binary classification and output labels (real jump / foam pseudo jump).

[0110] S340: Introducing Reynolds number and shear velocity correction mechanism: To further improve the stability of the model, the system introduces the Reynolds number related to the liquid motion state:

[0111]

[0112] in, is the liquid density; is the aspiration flow rate; is the channel diameter; is the liquid viscosity, and at the same time, calculate the local shear velocity:

[0113]

[0114] The shear rate is used to determine whether the foam structure has been densely populated. If a sudden change in high shear and low air pressure occurs, further confirmation is required to determine whether it is a pseudo-liquid surface.

[0115] S350: Construct velocity profile model and optimize interference identification strategy: Combined with channel radius , Reynolds number The segmented model is used to determine the liquid flow pattern according to the liquid motion state: When the flow is laminar, the jump point is stable and reliable; The turbulent flow and the jump point need to be confirmed in conjunction with vibration. The shear velocity fluctuates violently, and foam disturbance may occur. The flow state label output by the model is used to assist the interference determination module and improve the recognition confidence.

[0116] In this embodiment, steps S310-S350 can use the dual-modal information of air pressure signal + vibration signal to identify the interference source; quantitatively analyze the foam disturbance through physical parameter modeling (Re, shear velocity, etc.); construct a fusion discrimination model to significantly improve the anti-false judgment capability during the liquid surface recognition process; and adapt to complex liquid (such as foaming solutions, biological reagents, and gas-containing samples) scenarios.

[0117] The purpose of step S4 is to improve the stable recognition rate and structural analysis capability of multiple liquid surface contact points in the stratified liquid. Step S4 includes the following sub-steps S410-S450:

[0118] S410: Liquid level jump history record extraction: In the continuous liquid aspiration experiment, the system automatically records the jump time corresponding to each liquid level identification and compare it with the timestamp of the aspiration process , corresponding to the establishment of jump history data set:

[0119]

[0120] Each jump moment represents an identification event of the liquid surface layer, and this set serves as the basis of system learning samples.

[0121] S420: Construction of jump trajectory model: the historical jump time and its corresponding liquid absorption volume Mapping to trajectory points:

[0122]

[0123] in Indicates the The cumulative aspirated volume during each aspiration process; Sequence represents the time series record data of each jump behavior in the system history;

[0124] Use linear regression, polynomial fitting, or RNN / LSTM sequence modeling algorithms to fit the jump trajectory prediction model:

[0125]

[0126] Modeling functions using history , enter the volume of a certain aspiration , predict the time when the jump point may appear , for example, to build a function This can be achieved through the following linear regression model:

[0127]

[0128] This model can be used to dynamically predict the jump moment during the next liquid aspiration and generate candidate recognition windows in advance.

[0129] S430: Multi-jump point detection and depth inversion:

[0130] When multiple transition points are detected during the aspiration process When the system combines the liquid suction speed , instantaneous flow , aspiration time period The cumulative volume within , through the formula:

[0131]

[0132] Derived from the depth of the liquid surface layer from the starting point , which indicates the The position of the upper surface of the liquid layer.

[0133] S440: Hierarchical structure model generation and iterative optimization: When the system records multiple After that, a preliminary layered liquid structure is formed:

[0134]

[0135] Represents the stratified liquid level height sequence, which is the set of liquid level heights corresponding to multiple liquid level jump points identified in the liquid stratification detection system. Indicates the The height of the liquid level; : Indicates that the liquid levels are arranged in ascending direction, that is, they are arranged in monotonically increasing order from the bottom liquid level to the top liquid level.

[0136] The height sequence Usually obtained by the following steps:

[0137] Identify the jumping moment of the absorbing signal based on the jumping point detection algorithm (such as differential extreme value, slope change, curvature mutation, etc.);

[0138] Convert the jump time into the corresponding suction volume , combined with the volume / cross-sectional area parameters of the liquid suction system, the corresponding height of the liquid surface is further calculated , collect all identified liquid level heights into a sequence .

[0139] By calculating the depth distance between two adjacent layers , determine whether it conforms to the liquid density stratification hypothesis (density threshold and minimum layer height can be set);

[0140] For the case where the interlayer spacing is obviously abnormal (such as less than the minimum effective liquid layer height ), which can be eliminated or merged and corrected by sliding average or model residual filtering algorithm.

[0141] S450: Intelligent guidance for next jump point identification: Based on the trajectory model + current volume state, the system can predict the next jump point , and focus on monitoring the sudden change signal of air pressure near this time point to improve the success rate and efficiency of jump detection. The prediction window can be set to:

[0142]

[0143] in is the model's empirical error tolerance, such as 0.1s.

[0144] In this embodiment, steps S410-S450 establish a multi-layer liquid level jump database for the liquid suction system to form a trajectory prediction model; calculate the true height of each layer of liquid through inversion formulas and geometric modeling; automatically screen and correct abnormal jump points to improve the robustness of system recognition; and realize an intelligent layered recognition and prediction feedback mechanism for multi-layer liquid structures.

[0145] Step S5 is used to address the signal recognition instability problem caused by environmental disturbances, bubble interference or equipment errors, thereby improving the overall robustness and fault tolerance of the liquid level recognition system. It specifically includes the following sub-steps S510-S550:

[0146] S510: Identify abnormal state trigger conditions: During the liquid aspiration process, the system analyzes the collected air pressure curve in real time , slope , curvature Whether the characteristic indicators such as etc. conform to the preset jump mode;

[0147] The system triggers the fault tolerance mechanism when any of the following conditions occurs:

[0148] (a) The signal fluctuates violently, and it is impossible to extract obvious mutation points;

[0149] (b) The waveform oscillates repeatedly, misjudging multiple false transitions;

[0150] (c) The confidence level of the current recognition result (the probability value output by the AI ​​model) is lower than the tolerance threshold.

[0151] S520: Historical jump path callback mechanism: After the abnormal state is triggered, the system will first call the jump trajectory corresponding to similar aspiration conditions (such as aspiration speed, liquid type, tube diameter, etc.) in the historical data for comparison. Suppose the current aspiration flow rate is , search the history database for matching conditions:

[0152]

[0153] Indicates the flow rate of a historical sample With the current aspiration flow rate Compare the difference between them to determine whether the matching conditions are met and call the corresponding historical trajectory model , predict possible liquid level jump points or height estimation .

[0154] Historical trajectory model It refers to a liquid level jump moment prediction function model established for specific liquid types, operating speeds, and environmental conditions. Its input is the flow rate-volume characteristic sequence during the liquid aspiration process. , the output is the predicted jump time ;

[0155] Model training source: This model is derived from the inductive extraction and fitting of a large number of historical data samples. The training process is as follows: Raw data collection: Record the aspiration process of multiple experimental samples, where each group contains:

[0156]

[0157] Each is the velocity-volume time series of the liquid being sucked in, and is the actual liquid level jump moment marked, preferably based on a polynomial regression fitting model ; During actual detection, the system will perform the following steps:

[0158] Compare flow rate difference: If the current flow rate Compared with a historical sample flow rate The difference between ), indicating that the historical sample is similar to the current state.

[0159] Call the corresponding historical model: the system calls the prediction function associated with the sample , for the currently collected velocity-volume data sequence Take input and predict:

[0160]

[0161] That is, the time position of the liquid level jump point is predicted to set the recognition window in advance.

[0162] For example, if a simple linear model is used, then:

[0163]

[0164] in is the regression coefficient obtained by fitting the historical sample.

[0165] For example, if a neural network model is used, the structure is: input layer dimension = sequence length × feature dimension (such as flow rate, volume, timestamp); output layer dimension = 1 (i.e., predicted jump time ).

[0166] S530: Physical model path reasoning correction: If there is no valid match in historical data, the system will call the standard jump model built based on the principles of fluid mechanics, such as the ideal pressure mutation model, laminar flow rate compensation model, etc. to reverse the path. For example, the ideal pressure change model is as follows:

[0167]

[0168] For the moment pressure, is the initial pressure, For the moment The cumulative volume of liquid aspirated, is the proportionality coefficient;

[0169] The system is based on the currently measured pressure curve pressure data Fitting the above model , calculate the residual , find the time point with the smallest residual as the corrected jump position:

[0170]

[0171] This means that the system further corrects the jump point position through the minimum residual method near the time point predicted by the historical jump model to improve recognition accuracy and stability.

[0172] S540: Multi-layer criterion fusion based on model residuals: This combines the following correction information sources to construct the final liquid level recognition value:

[0173] The point where the model residual is minimum;

[0174] Jump trajectory model output;

[0175] Multimodal (air pressure + vibration) fusion results;

[0176] Depth limit constrained by the physical structure of the channel.

[0177] The final correction value is output using weighted fusion method:

[0178]

[0179] Indicates the candidate transition heights (predicted values ​​from different algorithms or models, such as edge detection method, residual method, model prediction method, etc.); Indicates the A high weight value is used to indicate its credibility or importance; Indicates the number of candidate heights; is the final corrected liquid level.

[0180] S550: Outputs a fault tolerance determination report and confidence level indicator. Specifically, the final system output includes: whether the current liquid level identification has been corrected; the correction method (historical trajectory / physical model / statistical average); the difference between the corrected value and the original identification value; and the confidence level of the correction result (low, medium, high). This visualization or numerical method provides feedback and optimization for subsequent aspiration steps.

[0181] In this embodiment, steps S510-S550 can still ensure the liquid level recognition capability under abnormal or unstable conditions of the air pressure signal; automatically call historical trajectory data or physical modeling path to complete predictive correction; support multi-source information fusion to improve the robustness and reliability of the overall recognition of the system; and avoid the impact of abnormal problems such as misidentification, missed identification, jump dislocation, etc. on the hierarchical structure judgment.

[0182] Furthermore, step S6 implements a technical solution for adaptively adjusting the model under different liquid types, detection scenarios, and environmental conditions. By monitoring system state parameters, the recognition threshold and parameter configuration are dynamically adjusted to achieve environmental adaptation and cross-scenario migration capabilities of the liquid level recognition model. This solution includes the following sub-steps S610-650:

[0183] Step S610: The system collects the current environment and operating status parameters in real time during operation, including but not limited to: ambient temperature , ambient air pressure , current liquid type label (by barcode or preset), detection channel radius , the current fluid Reynolds number , detecting sensor stability indicators (such as signal-to-noise ratio SNR).

[0184] The state parameter set is expressed as: ,in Indicates the type of liquid, such as water, oil, syrup, etc., and determines the fluid's density, viscosity and other physical properties.

[0185] S620: Model parameter sensitivity analysis: For each state parameter, the system constructs its sensitivity index to the recognition model based on historical recognition data. , indicating the effect of a parameter change on the liquid level judgment threshold Impact strength:

[0186]

[0187] The system prioritizes real-time correction and model compensation updates for highly sensitive parameters, including Output indicators of the liquid level recognition model (such as jump point position, liquid level height, recognition confidence, etc.); A set of state parameters A parameter in Indicates when When a small change occurs, the recognition output The intensity of the effect is called the “sensitivity index” of the parameter;

[0188] In step S620, the system constructs a model sensitivity analysis module based on historical recognition data to calculate the influence of each state parameter on the liquid level recognition output. The system can identify the key factors that are most sensitive to model changes, and give priority to their correction and dynamic compensation updates, thereby ensuring model stability and recognition accuracy, and improving the system's adaptability to various liquids and environmental changes.

[0189] S630: Adaptive determination threshold adjustment strategy: Based on the degree of environmental change, the system adjusts the jump recognition threshold in the following ways:

[0190]

[0191] in: Indicates the adaptive threshold for liquid level jump recognition in the current state; is the default recognition threshold in a standard environment (i.e., laboratory or calibration environment); Indicates the currently measured state parameter values ​​(such as temperature, pressure, Reynolds number, etc.), is the initial reference value of the corresponding state parameter; The empirical adjustment coefficient represents the influence weight of each state parameter change on the jump judgment (which can be obtained through training or fitting). Indicates the total number of state parameters.

[0192] During the liquid level jump point identification process, the system sets the current state parameters. Compared with the reference value under standard environment , combined with the influence coefficient of each parameter on the judgment threshold , calculate the adaptive jump recognition threshold ,The threshold can be adjusted in real time to adapt to various liquid characteristics and environmental fluctuations, thereby enhancing the recognition accuracy and stability of the system under complex working conditions.

[0193] S640: Model Structure Selection and Dynamic Migration Mechanism: The system automatically selects a pre-set model structure (e.g., single-layer neural network, decision tree, sliding window model, etc.) based on the current liquid properties (e.g., high viscosity, gas content, etc.) and the task type. For example, if the detected liquid is slow-flowing and high-viscosity (e.g., glycerin, oil), a curvature-based recognition model is selected; for low-viscosity aqueous solutions, a slope mutation rapid response model is selected; for low-SNR, a robust fusion model (e.g., decision forest) is activated. This mechanism dynamically adjusts the model structure to adapt to the scenario.

[0194] S650: Model migration and recalibration mechanism: If the system environment continuously changes beyond the scope of the original model (such as equipment redeployment in a new laboratory), the transfer learning mechanism can be called to re-fine-tune the model: use pre-loaded migration modules to extract key features; use a small amount of real jump data for incremental learning; adjust the decision boundary to converge to the optimal value for the new scenario.

[0195] This embodiment uses the adaptive optimization mechanism of steps S610-650 to perform real-time compensation for the liquid level recognition model based on external disturbances such as ambient temperature and pressure. It is highly adaptable to different liquid properties (density and viscosity). It can be deployed across devices and scenarios to improve versatility and migration capabilities. It ensures high recognition rate and stability even in complex or extreme environments.

[0196] Embodiment 2: This embodiment provides a layered liquid level detection system for realizing liquid level position recognition and layered structure estimation in a multi-layer liquid environment. The system includes:

[0197] Air pressure acquisition unit: It is configured inside the pipette pump or its connecting pipeline, and is used to collect real-time air pressure change data in the channel during the liquid aspiration process and output a high-frequency time-pressure sequence.

[0198] Signal processing and feature extraction unit: electrically connected to the air pressure acquisition unit, used to calculate the first-order derivative, second-order derivative and curvature of the pressure signal, and extract liquid surface contact-related features such as slope jump and inflection point change.

[0199] Liquid level jump identification unit: includes a historical sample modeling module and a dynamic threshold judgment module, which is used to identify liquid level mutation points within the estimated time window and verify the credibility of the results.

[0200] Foam disturbance identification module: used to analyze abnormal fluctuation signals caused by bubbles, and comprehensively judge and filter non-liquid surface jumps based on parameters such as vibration mode, amplitude change, and duration.

[0201] Reynolds number correction and fluid state determination module: used to collect channel flow velocity, liquid viscosity, density and other parameters, calculate the real-time Reynolds number, identify the laminar or turbulent flow state of the fluid, and correct the liquid level identification results.

[0202] Liquid level estimation module: Based on the identified jump point position, suction volume and channel structure parameters, combined with compression state modeling, the liquid level is calculated to achieve layered liquid level structure reconstruction.

[0203] Error tolerance and jump correction module: When anomalies occur or confidence is insufficient during the recognition process, this module calls the historical jump trajectory model and physical inversion model to perform jump point correction and error compensation.

[0204] Parameter adaptive adjustment module: used to collect environmental and operating status data in real time, dynamically update system thresholds and model structures, and maintain recognition stability across liquids and environments.

[0205] The system can be embedded in automated liquid handling equipment and is widely applicable to various application scenarios such as life sciences, precision chemicals, and food stratification detection, achieving high-precision and high-robustness multi-layer liquid surface intelligent recognition functions.

[0206] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters and thresholds in the formulas are set by technicians in this field according to actual conditions.

[0207] The above embodiments can be implemented in whole or in part via software, hardware, firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product comprises one or more computer instructions or computer programs. When loaded or executed on a computer, the processes or functions described in the embodiments of this application are fully or partially performed. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired means (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium accessible by a computer or a data storage device such as a server or data center that contains a collection of one or more available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, tapes), optical media (e.g., DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0208] Those skilled in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0209] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and modules described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0210] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the modules is only a logical function division. In actual implementation, there may be other division methods, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0211] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules, and may be located in one place or distributed across multiple network modules. Some or all of the modules may be selected to achieve the purpose of this embodiment according to actual needs.

[0212] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0213] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0214] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0215] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for detecting a stratified liquid level, characterized in that: The following steps are involved: S1. Collect the air pressure change data in the pipetting pump, combine the compression feedback characteristics, construct the air pressure compression state composite curve, process and analyze the curve, and extract the air pressure change trend and jump signal characteristics; S2. Establish a historical sample database based on the pressure changes and jump moments during the historical aspiration process; construct a dynamic threshold prediction model based on historical data and the aspiration process time series to generate a time judgment window for liquid level jumps; combine the time dimension of the pipetting process with the aspirated volume, link the channel structure parameters, and calculate the real-time liquid level height through a residual-optimized volume height fitting model; S3. Fusion of the air pressure change signal and the micro-vibration signal to identify whether there is signal interference caused by bubbles or foam; introduction of the current fluid's Reynolds number, shear rate, and pipeline structure characteristics to construct a flow velocity profile model; S4. Establish a liquid surface trajectory history database and train a trajectory prediction model to guide the determination of liquid surface jumps during the aspiration process. When multiple jump points are detected, multiple liquid surface contact positions are identified, and the corresponding depths of each liquid layer are estimated to complete the layer-by-layer identification of the stratified liquid. S1 specifically includes: collecting the air pressure change data during the liquid aspiration process to form a pressure time series; constructing a compression model of the relationship between air pressure and volume to reflect the air column compression behavior; Based on the pressure time series, the first-order derivative, second-order derivative and curvature value are calculated to extract the pressure change trend and jump characteristics. The extracted features are constructed into a standard signal vector as a candidate basis for liquid surface contact point identification.

2. A stratified liquid level detection method according to claim 1, characterized in that: It also includes S5. During the liquid aspiration process, if it is determined that the current signal has abnormal fluctuations or insufficient confidence, the reasoning path based on the historical trajectory or fluid physical model is automatically called to correct the liquid level recognition result.

3. A stratified liquid level detection method according to claim 2, characterized in that: It also includes S6, which automatically adjusts the prediction model, judgment threshold and sensing parameters according to changes in the system environmental status to adapt to different liquid properties and detection environments.

4. The method for detecting a stratified liquid level according to claim 1, wherein: S2 is specifically: The dynamic prediction model is trained based on the samples to determine the time window when the liquid level jump may occur; In this judgment window, the suction volume is calculated based on the current suction speed, and the liquid level height is deduced based on the air column structure; An error residual optimization function is established, and the credibility of liquid level determination is verified by the deviation between the model height and the volume inversion height.

5. The method for detecting a stratified liquid level according to claim 1, wherein: S3 specifically: The air pressure signal and the micro-vibration signal are used to construct a joint feature vector, and the interference recognition model is trained based on the feature vector; By identifying the bubble disturbance characteristics, the difference between the bubble desorption signal and the actual liquid level jump signal, the bubble interference can be judged and eliminated; The Reynolds number and shear rate are calculated based on the liquid density, channel diameter, flow velocity and viscosity. The liquid motion state is then used to identify whether it is in a laminar or turbulent state, and a segmented identification rule is established.

6. The method for detecting a stratified liquid level according to claim 1, wherein: S4 is specifically A jump trajectory prediction model is constructed based on historical jump records and aspirated volume, and the next jump moment is predicted in advance through trajectory modeling; When multiple jump points are detected, the liquid level depth corresponding to each jump point is calculated by combining the liquid aspiration speed and the instantaneous flow rate to construct a preliminary layered structure; Automatically screen or correct layers with abnormal depth spacing, and identify the next jump point by combining jump prediction and volume state to achieve intelligent identification of multi-layer liquid surface structure and dynamic inversion correction of jump points.

7. The method for detecting a stratified liquid level according to claim 2, wherein: S5 is specifically: Based on the real-time pressure curve characteristics, the abnormal state of the jump is judged and the fault tolerance mechanism is triggered; Prioritize matching historical liquid suction trajectories under abnormal conditions, and use the jump trajectory model under similar conditions to predict the liquid level jump point; If the historical matching is invalid, a standard jump path is constructed based on the physical model, and the minimum residual value is calculated to correct the jump point; Final liquid level recognition and judgment based on multi-source feature fusion; Output the recognition confidence level and whether to correct the mark, and use the correction result to optimize the subsequent recognition process.

8. The method for detecting a stratified liquid level according to claim 3, wherein: Step S6 is specifically as follows: Real-time collection of environmental status parameters and operating status data; Analyze the sensitivity of the model recognition threshold to each state parameter, and make real-time compensation for key parameters or modify model weights based on the sensitivity results; Adaptively adjust the recognition judgment threshold based on the degree of change of state parameters to achieve dynamic update of the threshold as the environment drifts; Automatically select the most suitable model structure type based on liquid properties and task requirements; When the system environment changes continuously beyond the original applicable range of the model, the model structure migration and parameter recalibration are triggered to perform key feature extraction and reinforcement learning.

Images