A multi-sensor liquid-liquid two-phase interface detection method and system

Abstract

The application provides a multi-sensor liquid-liquid two-phase interface detection method and system, and relates to the technical field of liquid-liquid two-phase interface detection. The method comprises the following steps: acquiring multi-source and multi-height interface physical property data, and pre-processing the data to obtain a sensing element signal curve; establishing a multi-sensor response model library for different interface states, wherein the response model library comprises a response curve of a typical signal of each sensing element under different interface transition layer states and a corresponding interface state description; acquiring a response curve of each sensing element in real time; adjusting the contribution weight of each sensing element in interface positioning according to the response curve and the corresponding characteristic curve in the response model library; and determining a core position or a characteristic area of the interface according to the adjusted contribution weight and a preset interface positioning rule. The multi-sensor liquid-liquid two-phase interface detection method and system provided by the application significantly improves the accuracy of liquid-liquid two-phase interface detection.

Description

A multi-sensor method and system for detecting liquid-liquid two-phase interfaces Technical Field

[0001] This application relates to the field of liquid-liquid interface detection technology, and more specifically, to a multi-sensor liquid-liquid interface detection method and system. Background Technology

[0002] In industrial production processes, accurately detecting and controlling the interface between two immiscible liquids is crucial. Modern industrial systems typically deploy multi-sensor systems composed of various sensing elements to obtain more comprehensive and accurate interface information. However, when the interface is no longer a clear boundary but evolves into a transitional layer with gradually changing physical properties, sensing elements based on different principles exhibit drastically different response characteristics, leading to highly inconsistent output signals. Existing information integration logic struggles to effectively handle these contradictory signals, while experience-based parameter adjustments often fail to address the root cause and may even introduce new risks, causing system performance to degrade under process disturbances, thus making it difficult to accurately identify and locate the true interface position. Summary of the Invention

[0003] This application aims to address at least one of the technical problems existing in the prior art. To this end, this application proposes a multi-sensor liquid-liquid interface detection method and system, aiming to improve the accuracy of liquid-liquid interface detection.

[0004] In a first aspect, the multi-sensor liquid-liquid interface detection method provided in the embodiments of this application includes:

[0005] Acquire multi-source, multi-height interface physical property data, and preprocess the data to obtain the signal curve of the sensing element;

[0006] Establish a response model library for multiple sensors to different interface states. The response model library contains response curves of typical signals of each sensing element under different interface transition layer states and corresponding interface state descriptions.

[0007] Real-time acquisition of the response curve of each sensing element;

[0008] Based on the response curve and the corresponding feature curve in the response model library, adjust the contribution weight of each sensing element in interface positioning.

[0009] Based on the adjusted contribution weights and preset interface positioning rules, the core location or feature area of ​​the interface is determined.

[0010] According to some embodiments of this application, the step of adjusting the contribution weight of each sensing element in interface positioning includes:

[0011] The real-time response curve of each sensing element is transformed into a multi-dimensional real-time feature vector, which includes the curve's shape descriptor, gradient features, local fluctuation features, and curve complexity index.

[0012] Calculate the Euclidean distance between each of the real-time feature vectors and the feature vectors of all known interface states in the response model library;

[0013] Calculate the unknown score based on the Euclidean distance;

[0014] When the unknown score exceeds a preset threshold, the current interface state is marked as an abnormal unknown state;

[0015] When the current interface state is marked as an abnormal unknown state, the abnormal physical characteristics are identified and quantified based on the specific feature that contributes the most to the high deviation in the real-time feature vector.

[0016] Based on the aforementioned abnormal physical characteristics, the contribution weight of each sensing element in interface positioning is adjusted.

[0017] According to some embodiments of this application, the step of adjusting the contribution weight of each sensing element in interface positioning based on the abnormal physical characteristics includes:

[0018] Monitor the real-time signal quality indicators of each sensing element, including the signal-to-noise ratio, signal stability, and the degree of deviation of the signal from the background noise;

[0019] Based on the type and intensity of the abnormal physical features, and the real-time signal quality index, assess the degree of impact of the abnormal physical features on the signal reliability of each sensing element;

[0020] Based on the degree of impact on reliability, the contribution weight of each sensing element in interface positioning is adjusted.

[0021] According to some embodiments of this application, the step of evaluating the impact of the abnormal physical features on the signal reliability of each sensing element based on the type and intensity of the abnormal physical features and the real-time signal quality index includes:

[0022] Identify multiple simultaneous and interacting anomalous physical features;

[0023] The sensitivity differences of the measurement principle of each sensing element to different combinations of abnormal physical characteristics were analyzed.

[0024] A multi-dimensional influence factor matrix is ​​established based on the anomalous physical characteristics of the interaction and the sensitivity differences.

[0025] By combining the influence factor matrix and the signal quality index, the comprehensive influence weight is dynamically calculated;

[0026] The impact of the abnormal physical characteristics on the signal reliability of each sensing element is evaluated based on the comprehensive influence weight.

[0027] According to some embodiments of this application, the step of adjusting the contribution weight of each sensing element in interface positioning based on the degree of reliability impact includes:

[0028] Calculate the fluctuation range of signal quality indicators for each sensing element;

[0029] Based on the fluctuation amplitude of the signal quality index, adjust the adjustment rate and adjustment step size of the contribution weight of each sensing element;

[0030] When the signal quality index fluctuates significantly, the monitoring frequency of the signal quality of the sensing element should be increased.

[0031] When the signal quality index continuously exceeds the preset signal range, the adjustment range of the contribution weight is limited.

[0032] According to some embodiments of this application, the determination step of the signal quality index continuously exceeding the preset signal range includes:

[0033] Calculate the average value and standard deviation of the signal quality index within a preset time window;

[0034] The adjusted confidence interval is obtained by adjusting the upper and lower limits of the confidence interval based on the mean and the standard deviation.

[0035] When multiple consecutive sampling points of the signal quality index are outside the adjusted confidence interval, and the deviation direction of the signal quality index is consistent, it is determined that the signal quality index continuously exceeds the preset confidence interval.

[0036] According to some embodiments of this application, the step of limiting the adjustment range of the contribution weight when the signal quality index continuously exceeds a preset signal range further includes:

[0037] When the signal quality index deviates from the adjustment confidence interval for more than a preset time length, multiple key process parameters are continuously monitored.

[0038] When any one or more of the key process parameters recover to the normal range within a preset time, and the signal quality indicators of each sensing element continue to recover to the confidence interval within the preset time, the current weight adjustment is stopped; the contribution weight of each sensing element is restored to the state before adjustment or the weight is re-evaluated and adjusted according to the recovered signal quality.

[0039] According to some embodiments of this application, the step of suspending the current weight adjustment when any one or more of the key process parameters recover to the normal range within a preset time, and the signal quality indicators of each sensing element continuously recover to the confidence interval within the preset time, includes:

[0040] When any one or more of the key process parameters recover to the normal range within a preset time, and the signal quality indicators of each sensing element continue to recover to the confidence interval within the preset time, the system enters the observation period:

[0041] During this observation period, the system continuously monitors the stability and trends of the key process parameters;

[0042] During this observation period, the system continuously monitors the stability and trend of the signal quality indicators;

[0043] When the key process parameters and the signal quality indicators remain stable and without abnormal fluctuations during the observation period, the current weight adjustment is terminated.

[0044] According to some embodiments of this application, the step of restoring the contribution weights of each sensing element to their pre-adjustment state or re-evaluating and adjusting the weights based on the restored signal quality includes:

[0045] Continuously monitor the real-time signal quality indicators of each sensing element;

[0046] Based on the measurement principle of each sensing element and its inherent response characteristics to the recovery state, determine the weighted recovery curve or recovery rate of each sensing element;

[0047] The contribution weight of each sensing element is asynchronously recovered based on the weight recovery curve or recovery rate of each sensing element.

[0048] When the signal quality index of the sensing element stabilizes and returns to the normal range, the weight of the sensing element is restored to its pre-adjustment state or the weight is re-evaluated and adjusted based on the restored signal quality.

[0049] Secondly, embodiments of this application provide a multi-sensor liquid-liquid interface detection system, including:

[0050] The raw data acquisition module is used to acquire interface physical property data from multiple sources and at multiple heights, and to preprocess the data to obtain the signal curve of the sensing element;

[0051] The model library building module is used to build a response model library for multiple sensors to different interface states. The response model library contains the response curves of typical signals of each sensing element in different interface transition layer states and the corresponding interface state descriptions.

[0052] The real-time data acquisition module is used to acquire the response curve of each sensing element in real time.

[0053] The weight adjustment module is used to adjust the contribution weight of each sensing element in interface positioning according to the response curve and the corresponding feature curve in the response model library.

[0054] The positioning output module is used to determine the core position or feature area of ​​the interface based on the adjusted contribution weight and the preset interface positioning rules.

[0055] The technical solution according to the embodiments of this application has at least the following beneficial effects:

[0056] This method effectively solves the problem in existing technologies where, when the liquid-liquid interface evolves into a transition layer with gradually changing physical properties, the output signals of sensing elements based on different principles are highly inconsistent, leading to difficulties in information integration and accurate identification and localization of the true interface location. By establishing a refined response model library, this application can more accurately understand the response characteristics of each sensing element under complex interface conditions. More importantly, by dynamically adjusting the contribution weight of each sensing element, this application can intelligently allocate the importance of each sensing element in interface localization based on the reliability of real-time signals and their matching degree with the model library, thereby effectively handling contradictory signals and avoiding the risks that may be introduced by traditional experience-based parameter adjustments. This allows the system to maintain excellent performance under process disturbances, significantly improving the accuracy and robustness of liquid-liquid interface detection, and providing reliable technical support for precise control in industrial production processes.

[0057] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0058] The accompanying drawings are used to provide a further understanding of the technical solutions of this application and constitute a part of the specification. They are used together with the embodiments of this application to explain the technical solutions of this application and do not constitute a limitation on the technical solutions of this application.

[0059] Figure 1 is a flowchart illustrating a multi-sensor liquid-liquid interface detection method provided in an embodiment of this application;

[0060] Figure 2 is a schematic diagram of a multi-sensor liquid-liquid interface detection system provided in one embodiment of this application. Detailed Implementation

[0061] To make the objectives, technical methods, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0062] Based on the above, this application proposes a multi-sensor liquid-liquid interface detection method and system, aiming to improve the accuracy of multi-sensor liquid-liquid interface detection.

[0063] Referring to Figure 1, Figure 1 is a schematic flowchart of a multi-sensor liquid-liquid interface detection method provided in an embodiment of this application. The multi-sensor liquid-liquid interface detection method provided in this embodiment includes, but is not limited to, steps S110 to S150, which will be described in detail below.

[0064] S110. Acquire multi-source, multi-height interface physical property data, and preprocess the data to obtain the signal curve of the sensing element;

[0065] S120. Establish a response model library for multiple sensors to different interface states. The response model library contains the response curves of typical signals of each sensing element under different interface transition layer states and the corresponding interface state descriptions.

[0066] S130. Real-time acquisition of the response curve of each sensing element;

[0067] S140. Adjust the contribution weight of each sensing element in interface positioning according to the response curve and the corresponding feature curve in the response model library.

[0068] S150. Based on the adjusted contribution weight and preset interface positioning rules, determine the core position or feature area of ​​the interface and output the information.

[0069] To make the technical solution of this application easier and clearer to understand, some key terms involved will be explained first.

[0070] "Multi-source, multi-height interface physical property data" refers to physical parameter information about the liquid-liquid interface region obtained from different types of sensing elements (such as capacitive, ultrasonic, and optical sensing) and at different vertical heights. This data can include dielectric constant, sound velocity, optical refractive index, density, etc., reflecting the physicochemical properties of the interface transition layer.

[0071] "Sensing element signal curve" refers to the sequence of physical quantity readings obtained by scanning each sensing element in the vertical direction or changing it over time at a fixed position. These readings can reflect the interface position and the characteristics of the transition layer.

[0072] The "Response Model Library" is a pre-built database that stores typical signal response curves of various sensing elements and their corresponding interface state descriptions under different known interface states (such as clear interfaces, emulsion layers, foam layers, etc.). This model library is the foundation for real-time signal comparison and state recognition.

[0073] "Interfacial transition layer state" refers to the fact that the liquid-liquid two-phase interface is not an ideal geometric plane, but a region with a certain thickness and continuously changing physical properties. Different process conditions may lead to different thicknesses, gradients, and compositions of the transition layer, which constitute different interfacial transition layer states.

[0074] "Contribution weight" refers to the relative importance or influence assigned to the signal of each sensing element when determining the final interface position. The greater the weight, the greater the influence of the signal of that sensing element on the final result.

[0075] "Interface positioning rules" refer to algorithms or logic used to determine the core location or feature region of an interface by integrating signals from multiple sensing elements and their contribution weights. These rules can be based on methods such as weighted averaging, fuzzy logic, and machine learning.

[0076] "Core location or characteristic region" refers to a single location point (core location) or a region (characteristic region) that is determined by the algorithm to be the most representative of the liquid-liquid interface.

[0077] This application provides a multi-sensor method for detecting liquid-liquid two-phase interfaces, the specific implementation of which is as follows:

[0078] The process of acquiring interface physical property data from multiple sources and at multiple heights, and preprocessing the data to obtain sensing element signal curves, can be implemented in various ways. For example, raw voltage, frequency, or time data collected from different sensing elements (such as capacitive sensors, ultrasonic sensors, optical sensors, etc.) at different heights can be manually input into the system. Subsequently, through manual inspection and correction, obvious noise and outliers are removed, and simple linear interpolation or smoothing is performed to obtain preliminary sensing element signal curves. Alternatively, a data acquisition system can automatically acquire data from multiple sensing elements and transmit the data to a processing unit via a preset communication protocol. The preprocessing process can include digital filtering (e.g., low-pass filtering to remove high-frequency noise), baseline correction, unit conversion, and data normalization to ensure the comparability of data of different types and dimensions and to generate standardized sensing element signal curves. For example, in an oil-water separator, multiple capacitive level gauges and ultrasonic level gauges can be installed vertically. The capacitive level gauge measures the change in dielectric constant, while the ultrasonic level gauge measures sound velocity or echo intensity. The system periodically reads raw data from these sensing elements and uses software algorithms to denoise and calibrate this raw data, ultimately generating signal curves that reflect the changes in dielectric constant and sound velocity with altitude.

[0079] In establishing a multi-sensor response model library for different interface states, the following methods can be adopted. For example, based on experience or literature review, several typical interface states (such as clear interface, mild emulsification, severe emulsification, etc.) can be manually defined, and a theoretical signal response curve for the sensing element can be assigned to each state. These curves can be simple mathematical functions, such as step functions or sigmoid functions. Simultaneously, a brief description of the interface state should be attached to each curve. Alternatively, various typical interface transition layer states can be simulated under controlled experimental conditions, and a large amount of actual data can be collected using a multi-sensor system. Using this data, typical signal response curves for each sensing element under different interface transition layer states can be identified through data analysis and pattern recognition techniques (e.g., cluster analysis, principal component analysis, etc.). These curves can be average curves, median curves, or clusters of curves with statistical characteristics. Simultaneously, the physicochemical parameters and visual observation results for each interface state should be recorded in detail as corresponding interface state descriptions and stored in a database to form a response model library. For example, in a laboratory, different degrees of oil-water emulsification can be simulated. For each emulsification state, measurements were taken using capacitive and ultrasonic sensors, and their signal curves were recorded. Statistical analysis of these experimental data yielded typical signal curves for the capacitive and ultrasonic sensors under three states: clear interface, mild emulsification, and severe emulsification. These curves, along with corresponding descriptions of emulsification degree, were then stored in a model library.

[0080] In the step of acquiring the response curve of each sensing element in real time, the following methods can be used. For example, the current output value of each sensing element can be manually read periodically and recorded. Then, these discrete points are connected to form a simple real-time response curve. As another implementation, the system can be configured with a high-speed data acquisition module to continuously acquire data from each sensing element at a preset sampling frequency. After preliminary filtering and formatting, these real-time data streams form continuous sensing element response curves that reflect the current interface state. These curves can be displayed on the operating interface in real time and used as input for subsequent analysis. For example, in an actual operating industrial storage tank, the system acquires hundreds of data points per second from each capacitive and ultrasonic sensor. These data points are integrated in real time to form the current capacitance signal curve and ultrasonic signal curve, which reflect the real-time state of the liquid-liquid interface in the storage tank.

[0081] In the step of adjusting the contribution weight of each sensing element in interface localization based on the response curve and the corresponding feature curve in the response model library, the following methods can be adopted. For example, an operator can subjectively judge the current interface state based on the visual similarity between the real-time response curve and the feature curve in the model library, and manually adjust the weight of each sensing element. For example, if the real-time capacitance curve looks closer to the "clear interface" curve in the model library, the weight of the capacitance sensor can be increased. As another implementation, the system can use pattern matching algorithms (e.g., correlation coefficient calculation, dynamic time warping (DTW) algorithm, or neural networks) to quantify the similarity between the real-time response curve and each feature curve in the model library. Based on the similarity level, the system can automatically evaluate the degree of matching between the current interface state and the known states in the model library. When the real-time curve highly matches the feature curve of a specific interface state, the system can increase the weight of sensing elements that perform more reliably in that state and decrease the weight of sensing elements that are more affected by that state, according to preset rules. For example, if the real-time capacitance curve is highly similar to the capacitance characteristic curve of the "emulsion layer" state in the model library, while the real-time ultrasonic curve is less similar to the ultrasonic characteristic curve of the "emulsion layer" state (possibly because the ultrasonic waves are severely attenuated in the emulsion layer), the system can automatically reduce the weight of the ultrasonic sensor in interface positioning and increase the weight of the capacitance sensor.

[0082] In the step of determining the core position or characteristic region of the interface based on the adjusted contribution weights and preset interface positioning rules, the following methods can be adopted. For example, the operator can manually perform a weighted average of the interface positions reported by each sensing element according to the adjusted weights to obtain a final interface position. Then, the operator verbally informs or manually records this position information. As another implementation method, the system can automatically calculate and determine the core position (a precise value) or a characteristic region with a certain range of the liquid-liquid two-phase interface based on the adjusted contribution weights and preset interface positioning rules (e.g., weighted average algorithm, fuzzy inference system, or machine learning-based decision model). For example, if the capacitive sensor has a weight of 0.7 and the ultrasonic sensor has a weight of 0.3, and they report the interface at 100mm and 110mm respectively, then the weighted average core position is 103mm. The system can output this core position or characteristic region information to the operator or upper control system through digital display, graphical interface, alarm signal, or data interface (e.g., OPC UA, Modbus). For example, in the extraction tower of petrochemical industry, the system weights and fuses the signals of multiple sensing elements according to the dynamically adjusted weights, and finally determines that the interface between the extraction phase and the refining phase is located at a certain precise position in the height of the tower. This position information is then sent to the DCS (Distributed Control System) in real time for subsequent flow or pump speed adjustments.

[0083] This dynamic, adaptive weight adjustment mechanism enables the method of this application to effectively overcome the limitations of traditional methods under complex interface conditions. It no longer relies on a single, fixed information integration logic, but rather intelligently optimizes the information fusion strategy according to changes in actual operating conditions. Therefore, this application can significantly improve the accuracy and robustness of liquid-liquid interface detection, especially when facing complex situations common in industrial production such as interface ambiguity, emulsification, and foaming, providing more reliable interface location information, thus providing strong support for precise control of industrial processes.

[0084] In one embodiment of this application, the contribution weight of each sensing element in interface positioning in step S140 is adjusted, including but not limited to steps S210 to S260. Each step will be described in turn below.

[0085] S210. Transform the real-time response curve of each sensing element into a multi-dimensional real-time feature vector. The real-time feature vector includes the curve's shape descriptor, gradient features, local fluctuation features, and curve complexity index.

[0086] S220. Calculate the Euclidean distance between each real-time feature vector and the feature vectors of all known interface states in the response model library.

[0087] S230. Calculate the unknown score based on the Euclidean distance;

[0088] S240. When the unknown score exceeds the preset threshold, mark the current interface state as an abnormal unknown state.

[0089] S250. When the current interface state is marked as an abnormal unknown state, identify and quantify the abnormal physical characteristics based on the specific feature that contributes the most to the high deviation in the real-time feature vector.

[0090] S260. Adjust the contribution weight of each sensing element in interface positioning based on abnormal physical characteristics.

[0091] Specifically, the real-time feature vector can be understood as a set of numerical values ​​used to describe specific attributes of the curve. For example, the curve's shape descriptor may include peak values, valley values, curve symmetry, skewness, etc., used to capture the overall morphological characteristics of the curve; gradient features may include first derivatives, second derivatives, etc., used to reflect the rate and acceleration of signal changes; local fluctuation features may include local variance, wavelet coefficients, etc., used to characterize the irregularity or noise level of the signal in a local region; curve complexity indicators may include fractal dimension, information entropy, etc., used to measure the complexity and information content of the curve. Through these multi-dimensional features, subtle changes in the interface state can be more precisely characterized.

[0092] Euclidean distance is a commonly used distance metric that can intuitively reflect the closeness of two vectors in a multidimensional space. By calculating the Euclidean distance between the real-time feature vector and the feature vector of each known interface state in the model library, the degree of deviation between the real-time state and all known states can be obtained.

[0093] Furthermore, the unknown score can be calculated based on the minimum, average, or specific statistics of the Euclidean distance. For example, when the Euclidean distance between the real-time feature vector and all known model feature vectors is large, the unknown score will increase accordingly, indicating that the current state is significantly different from any known state in the model library.

[0094] The preset threshold is a critical value set based on historical data or expert experience. When the unknown score exceeds this threshold, it means that the current interface may have physical phenomena or abnormal working conditions not covered by the model library, which require special handling.

[0095] When the current interface state is marked as an abnormal unknown state, the abnormal physical characteristics are identified and quantified based on the specific features that contribute the most to high deviation in the real-time feature vector. The purpose is to deeply analyze the essence of the abnormal state. Specifically, the main physical factors causing the anomaly can be inferred by analyzing which dimensions in the real-time feature vector differ most from the known model feature vector. For example, if the gradient feature has the highest deviation, it may mean an abnormal interface change rate; if the local fluctuation feature has a high deviation, it may indicate a violent disturbance or emulsification phenomenon at the interface. In this way, abnormal physical characteristics can be identified and quantified from both qualitative and quantitative perspectives.

[0096] Once the abnormal physical characteristics are identified, the weight of different sensing elements in interface localization can be dynamically adjusted based on their sensitivity to or resistance to interference with these characteristics. For example, if an emulsion layer is identified, and a certain sensing element is particularly sensitive to and susceptible to interference with the emulsion layer, its weight can be appropriately reduced; conversely, if a certain sensing element can still provide reliable information in the presence of an emulsion layer, its weight can be maintained or increased.

[0097] The proposed solution transforms the real-time response curve of the sensing element into a multi-dimensional real-time feature vector and calculates its Euclidean distance to known states in the response model library to obtain an unknown score. When this score exceeds a preset threshold, the system can proactively identify abnormal unknown states and further analyze the specific features in the real-time feature vector that contribute the most to high deviations, thereby identifying and quantifying abnormal physical features. It is precisely because of this deep understanding of the essence of abnormal states that subsequent weight adjustments are no longer blindly based on model matching, but can be tailored to the type and intensity of abnormal physical features, specifically adjusting the contribution weights of each sensing element. This effectively solves the problems of inaccurate weight adjustment and decreased positioning accuracy in traditional methods when facing complex or unknown interface states.

[0098] In one embodiment of this application, regarding step S260, which adjusts the contribution weight of each sensing element in interface positioning based on abnormal physical characteristics, including but not limited to steps S310 to S330, this step is described below.

[0099] S310. Monitor the real-time signal quality indicators of each sensing element. The signal quality indicators include the signal-to-noise ratio, signal stability, and the degree of deviation of the signal from the background noise.

[0100] S320. Based on the type and intensity of the abnormal physical characteristics and the real-time signal quality indicators, assess the degree of impact of the abnormal physical characteristics on the signal reliability of each sensing element.

[0101] S330. Adjust the contribution weight of each sensing element in interface positioning according to the degree of reliability impact.

[0102] Specifically, the real-time signal quality indicators refer to a set of parameters used to quantify the quality level of the signals of each sensing element under its current operating state. Among these, the signal-to-noise ratio (SNR) is the ratio of signal power to noise power, used to measure the relative strength of effective information and interference noise in the signal; signal stability refers to the degree of signal fluctuation over a period of time, reflecting the signal's smoothness; and the deviation of the signal from background noise refers to the difference between the signal amplitude or characteristics of the sensing element and the normal background noise level, used to indicate whether the signal is subject to abnormal interference or significant changes. Monitoring these indicators aims to obtain the real-time health status of the signals of each sensing element, with the purpose of providing a quantitative basis for subsequent reliability assessments.

[0103] Furthermore, the step of assessing the impact of the anomalous physical features on the signal reliability of each sensing element aims to comprehensively consider the inherent attributes of the identified anomalous physical features (such as type and intensity) and the current signal quality of each sensing element, in order to quantify the impact of these anomalous features on the reliability or availability of the signal output of each sensing element. For example, when a bubble anomalous physical feature is detected, an acoustic sensor may be significantly affected, with its signal-to-noise ratio decreasing significantly, while a capacitive sensor may be less affected. Through this assessment, a more refined understanding can be gained of which sensing element signals remain reliable under specific anomalous conditions, and which require a reduction in their weight.

[0104] Therefore, adjusting the contribution weight of each sensing element in interface localization based on the degree of reliability impact means dynamically adjusting its importance in the final interface localization decision based on the assessed reliability of each sensing element's signal. For example, if the signal reliability of a certain sensing element is assessed as low, its contribution weight will be reduced accordingly, or even temporarily excluded in extreme cases, to avoid its unreliable signal negatively impacting the overall localization result. Conversely, if the signal of a certain sensing element maintains high reliability under abnormal conditions, its contribution weight can be maintained or appropriately increased to fully utilize its effective information.

[0105] This application's solution addresses the problem of how to more accurately and robustly adjust the contribution weights of sensing elements when abnormal physical features are present by introducing real-time monitoring of the signal quality indicators of each sensing element and combining this with the type and intensity of identified abnormal physical features to evaluate the reliability of the signals from each sensing element. Because of the real-time monitoring of signal quality indicators (such as signal-to-noise ratio, stability, and deviation), the system can obtain the actual operating status of each sensing element in the current environment. Given that abnormal physical features may have different effects on different types of sensing elements, combining the type and intensity of abnormal physical features with real-time signal quality indicators allows for a more comprehensive and refined assessment of the actual reliability of each sensing element's signal. This reliability-based weight adjustment mechanism enables the system to dynamically and adaptively adjust the contribution weights of each sensing element, ensuring that, under complex and changing interface conditions, data from sensing elements with high signal quality and strong reliability are always prioritized, thereby effectively avoiding positioning errors caused by abnormal interference from individual sensing element signals.

[0106] In one embodiment of this application, step S320, which assesses the impact of abnormal physical features on the signal reliability of each sensing element based on the type and intensity of the abnormal physical features and the real-time signal quality index, includes, but is not limited to, steps S410 to S450. Each step will be described in turn below.

[0107] S410. Identify multiple simultaneous and interacting anomalous physical features;

[0108] S420. Analyze the sensitivity differences of the measurement principle of each sensing element to different combinations of abnormal physical characteristics;

[0109] S430. Establish a multi-dimensional influence factor matrix based on the abnormal physical characteristics and sensitivity differences of the interaction;

[0110] S440. Combine the influence factor matrix and signal quality indicators to dynamically calculate the comprehensive influence weight;

[0111] S450. Based on the comprehensive influence weight, assess the degree of impact of abnormal physical characteristics on the signal reliability of each sensing element.

[0112] Specifically, identifying multiple concurrent and interacting anomalous physical features means that when an abnormal unknown state is detected, the system not only identifies a single anomalous physical feature, such as an emulsion layer or bubbles, but also further analyzes whether multiple anomalous physical features occur simultaneously and influence each other. For example, an emulsion layer may contain microbubbles, or scale buildup on the sensor surface may be accompanied by localized temperature fluctuations. This step aims to extract and distinguish these complex combinations of anomalies from real-time feature vectors using advanced pattern recognition algorithms or expert systems.

[0113] Analyzing the sensitivity differences of each sensing element's measurement principle to different combinations of anomalous physical characteristics can be understood as follows: different sensing elements, such as capacitive, conductive, or optical sensors, have different response characteristics to specific combinations of anomalous physical characteristics. For example, a capacitive sensor may be highly sensitive to changes in emulsion layer thickness, while a conductive sensor may be more sensitive to changes in ion concentration within the emulsion layer, and an optical sensor may respond significantly to changes in turbidity or color. This step quantifies the responsiveness and reliability of each sensing element to different combinations of anomalous physical characteristics through a pre-established sensor characteristic database or through online calibration.

[0114] In practical applications, a multi-dimensional influence factor matrix is ​​established based on the abnormal physical characteristics of the interactions and the sensitivity differences. Specifically, a mathematical model is constructed that can characterize the degree of influence of various combinations of abnormal physical characteristics on the signal reliability of different sensing elements. The dimensions of this matrix can include the type, intensity, and interaction mode of the abnormal physical characteristics, as well as the sensitivity parameters of each sensing element. For example, an element in the matrix can represent the influence factor of the abnormal combination of "coexistence of emulsion layer and bubbles" on the signal reliability of a "capacitive sensor".

[0115] Furthermore, dynamically calculating the comprehensive influence weight by combining the aforementioned influence factor matrix and the signal quality indicators refers to fusing the theoretical or empirical influence levels provided by the influence factor matrix with the real-time monitored signal quality indicators (such as signal-to-noise ratio and stability). This process can employ methods such as weighted averaging, fuzzy logic reasoning, or machine learning models to adjust the comprehensive evaluation weight of the signal reliability of each sensing element under the current abnormal state in real time. This dynamic calculation ensures that the evaluation results can adapt to constantly changing operating conditions.

[0116] Therefore, based on the aforementioned comprehensive influence weights, the impact of the abnormal physical characteristics on the signal reliability of each sensing element is evaluated, with the aim of obtaining a more accurate and comprehensive reliability assessment result. This assessment result will be directly used to subsequently adjust the contribution weight of each sensing element in interface positioning, ensuring that under complex abnormal conditions, the system can prioritize the data from sensing elements that are less affected by abnormalities or have stronger robustness to abnormalities.

[0117] This application's solution overcomes the limitations of considering only a single anomalous feature by identifying multiple simultaneously occurring and interacting anomalous physical characteristics, enabling the system to more comprehensively understand the interface state under complex operating conditions. By analyzing the sensitivity differences of each sensing element's measurement principle to different combinations of anomalous physical characteristics, this application's solution can specifically evaluate the performance of different sensors under specific anomalous conditions, avoiding a "one-size-fits-all" reliability judgment. Based on this, a multi-dimensional influence factor matrix is ​​established to quantify complex interactions and sensitivity differences, providing a solid mathematical foundation for subsequent reliability assessment. Finally, by combining the aforementioned influence factor matrix and real-time signal quality indicators to dynamically calculate the comprehensive influence weight, the assessment of the influence of each sensing element's signal reliability becomes more accurate and real-time, thereby more effectively guiding the adjustment of the sensing element's contribution weight and ensuring the accuracy and robustness of interface positioning under anomalous operating conditions.

[0118] In one embodiment of this application, the above step S330 adjusts the contribution weight of each sensing element in interface positioning according to the degree of reliability impact, including but not limited to steps S510 to S540. Each step will be described in turn below.

[0119] S510, Calculate the fluctuation range of signal quality indicators for each sensing element;

[0120] S520. Adjust the adjustment rate and adjustment step size of the contribution weight of each sensing element according to the fluctuation range of the signal quality index.

[0121] S530. When the signal quality index fluctuates significantly, increase the monitoring frequency of the signal quality of the sensing element.

[0122] S540. When the signal quality index continuously exceeds the preset signal range, the adjustment range of the contribution weight is limited.

[0123] Specifically, calculating the fluctuation range of signal quality indicators for each sensing element refers to quantifying the degree of change in signal quality indicators over a period of time using statistical methods, such as calculating standard deviation, variance, or mean absolute deviation. The purpose is to monitor the stability of the signal in real time.

[0124] The adjustment rate and step size for the contribution weights of each sensing element, based on the fluctuation range of the signal quality index, can be understood as follows: when the signal quality fluctuation is small, a smaller adjustment rate and step size can be used to achieve smooth and precise weight adjustment; while when the fluctuation is large, it may be necessary to more cautiously reduce the adjustment rate and step size to avoid over-adjustment due to signal instability. The aim is to make the weight adjustment process more adaptable to real-time signal changes, improving the stability and accuracy of the adjustment.

[0125] In practical applications, when the signal quality index fluctuates significantly, increasing the monitoring frequency of the sensing element's signal quality means that the system will collect and analyze the sensing element's signal quality data more frequently in order to promptly detect and respond to further changes in the signal. For example, the monitoring frequency can be increased from once per second to once every 0.1 seconds. The purpose is to provide more intensive feedback when the signal is unstable, providing more timely data support for subsequent decision-making.

[0126] Furthermore, when the signal quality index continuously exceeds the preset signal range, limiting the adjustment range of the contribution weight means that once the signal quality of a certain sensing element is in an unreliable state for a long time (e.g., low signal-to-noise ratio, poor stability), the system will no longer allow its contribution weight to be adjusted significantly, and may even freeze its weight or set a low upper limit. The purpose is to prevent unreliable sensor signals from negatively impacting the overall interface positioning results and to maintain the overall robustness of the system.

[0127] Through the above technical solutions, this application can significantly improve the adaptability and robustness of the multi-sensor liquid-liquid interface detection method in dynamic and complex environments. Specifically, by dynamically adjusting the rate and step size of weight adjustment, the system can better cope with signal fluctuations and avoid positioning errors caused by over-adjustment or under-adjustment. Increasing the monitoring frequency ensures timely acquisition of signal status at critical moments, providing a basis for rapid response to abnormal situations. Especially when signal quality continues to deteriorate, limiting the adjustment range of contribution weights effectively isolates the interference of unreliable sensors on the overall decision-making, thereby ensuring the accuracy and reliability of interface positioning, reducing the risk of misjudgment, and improving the overall performance and stability of the system.

[0128] In one embodiment of this application, the judgment step regarding the signal quality index continuously exceeding the preset signal interval in step S540 includes, but is not limited to, steps S610 to S630, which will be described in turn below.

[0129] S610. Calculate the average value and standard deviation of the signal quality index within a preset time window;

[0130] S620. Adjust the upper and lower limits of the confidence interval based on the mean and standard deviation;

[0131] S630. When multiple consecutive sampling points of the signal quality index are outside the adjusted confidence interval, and the deviation direction of the signal quality index is consistent, it is determined that the signal quality index continuously exceeds the preset confidence interval.

[0132] Specifically, the signal quality index can be understood as a quantitative parameter reflecting the reliability of the sensing element's signal, such as signal-to-noise ratio, signal stability, or deviation from background noise. To accurately capture its dynamic characteristics, firstly, the signal quality index is statistically analyzed within a preset time window, calculating its average value and standard deviation. The length of the preset time window can be set according to the frequency of signal changes in the actual application scenario and the system's sensitivity requirements for anomaly responses. The average value reflects the central trend of the signal quality index, while the standard deviation characterizes its fluctuation degree.

[0133] Furthermore, based on the calculated mean and standard deviation, the upper and lower limits of the confidence interval are dynamically adjusted. This adjustment mechanism allows the confidence interval to adaptively reflect the actual distribution and fluctuation range of the current signal quality indicators, rather than using a fixed, static interval that may not be applicable to all operating conditions. For example, when the signal fluctuates significantly, the confidence interval will be appropriately widened to reduce false alarms; when the signal is stable, the confidence interval will be narrowed to improve detection sensitivity.

[0134] Based on this, the signal quality index is only determined to be continuously exceeding the preset confidence interval when multiple consecutive sampling points are outside the adjusted confidence interval and the deviation direction of the signal quality index is consistent. The requirement of "multiple consecutive sampling points" aims to filter out transient noise or occasional fluctuations, ensuring that the detected deviation is continuous and meaningful. The requirement of "consistent deviation direction" further enhances the reliability of the judgment. For example, if the signal quality index is continuously below the lower limit of the confidence interval, it indicates a continuous deterioration in signal quality; if it is continuously above the upper limit, it may indicate some kind of abnormal enhancement. These two conditions work together to avoid misjudgments caused by transient anomalies or bidirectional fluctuations.

[0135] Through the above technical solution, this application provides a more intelligent and robust mechanism for judging signal quality anomalies. Compared with simple fixed threshold judgment, this scheme can significantly reduce the false alarm rate and avoid incorrectly limiting the contribution weight of sensing elements due to instantaneous noise or normal fluctuations. Simultaneously, by dynamically adjusting the confidence interval, the system has stronger adaptability to signal characteristics under different operating conditions, ensuring timely and accurate identification when signal quality truly deteriorates continuously. This effectively limits the weight adjustment range of unreliable sensing elements, maintaining the overall accuracy and stability of liquid-liquid interface detection. This adaptive, multi-condition joint judgment method significantly improves the reliability and decision quality of the system in complex and variable environments.

[0136] In one embodiment of this application, the steps following step S540, which limits the adjustment range of the contribution weight when the signal quality index continuously exceeds the preset signal range, include, but are not limited to, steps S710 to S740. Each step will be described in turn below.

[0137] S710. When the signal quality index deviates from the adjustment confidence interval for more than the preset time length, multiple key process parameters are continuously monitored.

[0138] S720. When any one or more of the key process parameters recover to the normal range within a preset time, and the signal quality indicators of each sensing element continue to recover to the confidence interval within a preset time, the current weight adjustment is stopped; the contribution weight of each sensing element is restored to the state before adjustment or the weight is re-evaluated and adjusted according to the recovered signal quality.

[0139] Specifically, the "preset time length" can be set according to the dynamic characteristics of the actual process and the tolerance for abnormal responses; for example, it can be set to several minutes to several hours. Continuous monitoring of multiple key process parameters means that the system not only focuses on the signal quality of the sensing element itself, but also extends to real-time tracking of external process conditions that affect the state of the liquid-liquid interface and the performance of the sensing element. These key process parameters may include, but are not limited to, temperature, pressure, flow rate, component concentration, stirring speed, etc., with the aim of more comprehensively diagnosing the root cause of signal anomalies.

[0140] The "preset time" can be understood as the observation period for the system to determine whether process parameters and signal quality have stabilized and recovered. Its length should be sufficient to ensure the reliability of the recovery; for example, it can be set to several seconds to several minutes. When any one or more of the key process parameters recover to the normal range within the preset time, it means that the external process disturbance causing the signal abnormality has been controlled or eliminated. At the same time, if the signal quality indicators of each sensing element continuously recover to the confidence interval within the preset time, it indicates that the performance and signal output of the sensing element itself have returned to normal. When these two conditions are met, the system will stop the current weight adjustment, that is, stop the restriction or special adjustment strategies previously adopted due to signal abnormalities.

[0141] In practical applications, restoring the contribution weights of each sensing element to their pre-adjustment state means resetting the weights to the known good state before the anomaly occurred. This is suitable when the anomaly is a temporary disturbance and the system can fully recover. Alternatively, re-evaluating and adjusting the weights based on the recovered signal quality means that after the signal quality is restored, the system recalculates and optimizes the contribution weights of each sensing element based on the current actual signal quality indicators to ensure that the accuracy and robustness of the interface positioning are optimal in the new stable state.

[0142] This application's solution introduces continuous monitoring of key process parameters and combines this monitoring with the recovery status of signal quality indicators from sensing elements, constructing a more comprehensive anomaly recovery judgment mechanism. When signal quality indicators continuously deviate, the system no longer passively restricts weight adjustments but actively investigates the underlying causes of the anomaly. By monitoring key process parameters, it can be determined whether the signal anomaly is caused by external process fluctuations. Once both key process parameters and sensing element signal quality indicators recover to normal ranges within a preset time, it indicates that the system has effectively recovered from the abnormal state. At this point, suspending the current weight adjustment and restoring or re-evaluating the weights can prevent the system from remaining in a conservative or suboptimal weight configuration for an extended period, thereby ensuring the timeliness and accuracy of interface positioning. This mechanism enables the system to respond more intelligently to process changes, improving its adaptability and reliability in complex industrial environments.

[0143] In one embodiment of this application, the step of suspending the current weight adjustment in step S720, where any one or more of the key process parameters recover to the normal range within a preset time, and the signal quality indicators of each sensing element continue to recover to the confidence interval within the preset time, includes:

[0144] S810. When any one or more of the key process parameters recover to the normal range within a preset time, and the signal quality indicators of each sensing element continue to recover to the confidence interval within the preset time, the system enters the observation period:

[0145] S820. During this observation period, the system continuously monitors the stability and trends of key process parameters;

[0146] S830. During this observation period, the system continuously monitors the stability and trend of signal quality indicators;

[0147] S840. When the key process parameters and signal quality indicators remain stable and without abnormal fluctuations during the observation period, the current weight adjustment shall be stopped.

[0148] Specifically, the "observation period" refers to a pre-set monitoring phase that continues for a certain period of time after key process parameters and signal quality indicators have initially returned to normal ranges. The length of this observation period can be flexibly configured according to the actual application scenario, the fluctuation characteristics of process parameters, and the requirements for system stability; for example, it can be set to several minutes, several hours, or even longer. Its purpose is to provide a buffer time to verify the persistence and stability of the system recovery.

[0149] During the observation period, "continuously monitoring the stability and trend of the key process parameters" means that the system will continuously collect and analyze real-time data of the key process parameters, assess whether their values ​​fluctuate within the normal range, and whether their direction and rate of change meet expectations. For example, the moving average and standard deviation of the parameters can be calculated and compared with historical stable data to determine whether there is any abnormal drift or fluctuation.

[0150] Meanwhile, during the observation period, "continuously monitoring the stability and trend of the signal quality indicators" means that the system will continuously monitor the signal quality indicators of each sensing element, such as signal-to-noise ratio and stability, to ensure that these indicators not only recover to the confidence interval but also remain stable within that interval without any signs of further deterioration or drastic fluctuations. This helps to confirm that the signal output of the sensing elements has truly recovered to a reliable level.

[0151] In practical applications, "stop the current weight adjustment when the key process parameters and the signal quality indicators remain stable and without abnormal fluctuations during the observation period" means that the system will only finally confirm the reliability of the recovery state and execute the weight adjustment stop operation when all relevant key process parameters and sensing element signal quality indicators show continuous stability throughout the entire observation period and no abnormal fluctuations or deviations from the normal trend are detected.

[0152] Through the above technical solution, this application can significantly improve the robustness and reliability of the multi-sensor liquid-liquid interface detection method. Specifically, by introducing an observation period and continuously monitoring the stability and trends of key process parameters and signal quality indicators, the risk of prematurely terminating weight adjustments due to instantaneous recovery is avoided. This ensures that the accuracy and stability of interface positioning are continuously guaranteed after the system resumes normal operation, effectively preventing the impact of potential secondary anomalies or fluctuations on the detection results. This more prudent recovery strategy enables the system to provide more stable and reliable interface detection services in the face of complex and ever-changing industrial environments, thereby reducing the false positive rate and operational risks.

[0153] In one embodiment of this application, the above step S720, which restores the contribution weight of each sensing element to its state before adjustment or re-evaluates and adjusts the weight based on the restored signal quality, includes, but is not limited to, steps S910 to S940. Each step will be described in turn below.

[0154] S910: Continuously monitors the real-time signal quality indicators of each sensing element;

[0155] S920. Based on the measurement principle of each sensing element and its inherent response characteristics to the recovery state, determine the weighted recovery curve or recovery rate of each sensing element.

[0156] S930. Asynchronously recover the contribution weight of each sensing element according to the weight recovery curve or recovery rate of each sensing element;

[0157] S940. When the signal quality index of the sensing element stabilizes and returns to the normal range, restore the weight of the sensing element to its pre-adjustment state or re-evaluate and adjust the weight based on the restored signal quality.

[0158] Specifically, continuous monitoring of the real-time signal quality indicators of each sensing element refers to the system continuously acquiring and analyzing the signal quality data of each sensing element, such as the signal-to-noise ratio, signal stability, and the degree of deviation of the signal from background noise, in order to grasp its recovery status in real time. Its purpose is to provide accurate and timely basis for subsequent weighted recovery decisions.

[0159] In this process, based on the measurement principles and inherent response characteristics of each sensing element to the recovery state, the weighted recovery curve or recovery rate of each sensing element is determined. This can be understood as follows: for different types of sensing elements, such as capacitive sensors, ultrasonic sensors, or optical sensors, the change patterns and speeds of their signal quality indicators during recovery from an abnormal state to a normal state may differ significantly. Therefore, it is necessary to pre-establish or dynamically learn a specific weighted recovery curve or recovery rate model for each sensing element based on its physical operating principle (e.g., the response speed of a capacitive sensor to changes in dielectric constant, and the response speed of an ultrasonic sensor to changes in sound velocity) and its inherent recovery characteristics after experiencing abnormal disturbances (e.g., the settling time of the sensor's internal circuitry, and the response hysteresis of physical components). These curves or rate models describe how the contribution weight of the sensing element should be gradually adjusted during the gradual recovery of signal quality. The purpose is to ensure that the weight adjustment process matches the actual recovery state of the sensing element, avoiding premature or delayed weight recovery.

[0160] In practical applications, the contribution weights of each sensing element are asynchronously recovered based on their weight recovery curves or recovery rates. Specifically, when multiple sensing elements are recovering simultaneously, their contribution weights are not adjusted synchronously with the same step size or rate. Instead, the weight adjustment of each sensing element will independently follow its preset or dynamically determined recovery curve or rate.

[0161] Furthermore, when the signal quality index of a sensing element stabilizes and returns to the normal range, the weight of that sensing element is restored to its pre-adjustment state, or the weight is reassessed and adjusted based on the recovered signal quality. This means that when the real-time signal quality index of a sensing element, after continuous monitoring, is determined to have stabilized and returned to the preset normal operating range, the system will perform a final processing of the contribution weight of that sensing element. This could be by directly restoring the weight to the initial value before the anomaly occurred, or, to adapt to possible environmental changes or sensor aging, the system will reassess and adjust the weight based on the currently recovered signal quality index to ensure that its contribution weight is optimal. The purpose is to ensure that after the sensing element has fully recovered, it can play its due and accurate contribution in interface positioning.

[0162] This application's solution determines a unique weighted recovery curve or recovery rate by continuously monitoring the real-time signal quality indicators of each sensing element and combining this with the unique measurement principles and inherent response characteristics of each sensing element to the recovery state. This personalized recovery strategy enables asynchronous recovery of the contribution weights of each sensing element, thus avoiding the transient instability or accuracy degradation problems that may occur with traditional synchronous recovery. When the signal quality indicators of a sensing element stably recover to the normal range, its weight is restored to its pre-adjustment state or re-evaluated and adjusted based on the recovered signal quality. This ensures that the contribution weight of each sensing element during and after recovery matches its current actual performance, thereby maintaining the accuracy of interface positioning and the robustness of the system.

[0163] Referring to Figure 2, Figure 2 is a schematic diagram of a multi-sensor liquid-liquid interface detection system provided in one embodiment of this application. The multi-sensor liquid-liquid interface detection system 1000 includes:

[0164] The raw data acquisition module 1010 is used to acquire interface physical property data from multiple sources and at multiple heights, and to preprocess the data to obtain the signal curve of the sensing element.

[0165] The model library building module 1020 is used to build a response model library for multiple sensors to different interface states. The response model library contains the response curves of typical signals of each sensing element in different interface transition layer states and the corresponding interface state descriptions.

[0166] The real-time data acquisition module 1030 is used to acquire the response curve of each sensing element in real time;

[0167] The weight adjustment module 1040 is used to adjust the contribution weight of each sensing element in interface positioning according to the response curve and the corresponding feature curve in the response model library.

[0168] The positioning output module 1050 is used to determine the core position or feature area of ​​the interface based on the adjusted contribution weight and the preset interface positioning rules, and output the information.

[0169] Specifically, the raw data acquisition module is used to acquire interface physical property data from multiple sources and at multiple heights, and preprocess the data to obtain the signal curve of the sensing element. As one implementation, the raw data acquisition module can be configured to receive raw voltage, frequency, or time data from different types of sensing elements (e.g., capacitive sensors, ultrasonic sensors, optical sensors, etc.) via an analog input interface. This module may contain a simple signal conditioning circuit for amplifying, filtering, and other preliminary processing of the raw signal, converting it into a digital signal. Subsequently, through manual inspection and correction, obvious noise and outliers are removed, and simple linear interpolation or smoothing is performed to obtain the preliminary signal curve of the sensing element. As another preferred implementation, the raw data acquisition module can integrate a high-speed data acquisition card and a communication interface, enabling it to automatically acquire data from multiple sensing elements and transmit the data to the processing unit via a preset communication protocol (e.g., Modbus TCP / IP, Ethernet / IP, etc.). The preprocessing functions of this module can include digital filtering (e.g., low-pass filtering to remove high-frequency noise), baseline correction, unit conversion, and data normalization to ensure the comparability of data of different types and dimensions and to generate standardized sensing element signal curves.

[0170] The model library creation module provides a user interface that allows operators to manually input and define several typical interface states (such as clear interface, mild emulsification, severe emulsification, etc.) based on experience or literature review, and assign a theoretical sensing element signal response curve for each state. These curves can be simple mathematical functions, such as step functions or sigmoid functions. A brief description of the interface state is also provided for each curve. Alternatively, the model library creation module can integrate data analysis and pattern recognition algorithms. Under controlled experimental conditions, it simulates various typical interface transition layer states and collects a large amount of real-world data using a multi-sensor system. Using this data, techniques such as cluster analysis, principal component analysis, or deep learning are employed to automatically identify typical signal response curves for each sensing element under different interface transition layer states. These curves can be average curves, median curves, or clusters of curves with statistical characteristics. Simultaneously, this module can store detailed physicochemical parameters and visual observation results of the interface states as corresponding interface state descriptions, and store them in a database to form a response model library.

[0171] The real-time data acquisition module can be configured to periodically read and record the current output value of each sensing element via a serial communication interface or analog input port. These discrete points are then connected to form a simple real-time response curve. Alternatively, the real-time data acquisition module can be configured with a high-speed data acquisition module to continuously acquire data from each sensing element at a preset sampling frequency. After preliminary filtering and formatting, these real-time data streams form continuous sensing element response curves reflecting the current interface state. These curves can be displayed in real-time on the operating interface and used as input for subsequent analysis.

[0172] The weight adjustment module provides an interactive interface that allows operators to subjectively judge the current interface state based on the visual similarity between the real-time response curve and feature curves in the model library, and manually adjust the weight of each sensing element. For example, if the real-time capacitance curve looks closer to the "clear interface" curve in the model library, the weight of the capacitance sensor can be increased. As another preferred implementation, the weight adjustment module can integrate a pattern matching algorithm (e.g., correlation coefficient calculation, dynamic time warping (DTW) algorithm, or neural network) to automatically quantify the similarity between the real-time response curve and each feature curve in the model library. Based on the similarity level, the module can automatically assess the degree of matching between the current interface state and known states in the model library. When the real-time curve highly matches the feature curve of a specific interface state, the system can, according to preset rules, increase the weight of sensing elements that perform more reliably in that state and decrease the weight of sensing elements that are more affected by that state.

[0173] The positioning output module can provide a display interface where operators manually calculate a weighted average of the interface positions reported by each sensing element based on adjusted weights, thus obtaining a final interface position. The operator then verbally communicates or manually records this position information. Alternatively, in another preferred implementation, the positioning output module can integrate advanced algorithms to automatically calculate and determine the core position (a precise numerical value) or a characteristic region within a certain range of the liquid-liquid two-phase interface based on adjusted contribution weights and preset interface positioning rules (e.g., weighted average algorithm, fuzzy inference system, or machine learning-based decision model). This module can output this core position or characteristic region information to the operator or upper-level control system via digital display, graphical interface, alarm signal, or standard data interface (e.g., OPC UA, Modbus).

[0174] Therefore, the system of this application can significantly improve the accuracy and robustness of liquid-liquid interface detection, especially when facing complex situations such as interface ambiguity, emulsification, and foaming that are common in industrial production. It can provide more reliable interface location information, thereby providing strong support for the precise control of industrial processes.

Claims

1. A multi-sensor method for detecting liquid-liquid two-phase interfaces, characterized in that, Includes the following steps: The process involves acquiring multi-source, multi-height interface physical property data and preprocessing the data to obtain signal curves for sensing elements. This multi-source, multi-height interface physical property data refers to physical parameter information about the liquid-liquid two-phase interface region obtained from different types of sensing elements at different vertical heights. A multi-sensor response model library for different interface states is established, containing response curves of typical signals for each sensing element under different interface transition layer states, along with corresponding interface state descriptions. The response curve of each sensing element is acquired in real-time. The real-time response curve of each sensing element is converted into a multi-dimensional real-time feature vector, which includes a curve shape descriptor, gradient features, local fluctuation features, and curve complexity index. The Euclidean distance between each real-time feature vector and the feature vectors of all known interface states in the response model library is calculated. An unknown score is calculated based on the Euclidean distance. When the unknown score exceeds a preset threshold, the current interface state is marked as an abnormal unknown state. When the current interface state is marked as an abnormal unknown state, abnormal physical features are identified and quantified based on the specific features in the real-time feature vectors that contribute the most to high deviation. The system monitors the real-time signal quality indicators of each sensing element, including the signal-to-noise ratio, signal stability, and the degree of deviation between the signal and background noise. Based on the type and intensity of the abnormal physical features and the real-time signal quality indicators, the system assesses the impact of the abnormal physical features on the signal reliability of each sensing element. Based on the degree of reliability impact, the system adjusts the contribution weight of each sensing element in interface positioning. Based on the adjusted contribution weight and preset interface positioning rules, the system determines the core location or feature area of ​​the interface.

2. The multi-sensor liquid-liquid interface detection method according to claim 1, characterized in that, The step of assessing the impact of the abnormal physical features on the signal reliability of each sensing element based on the type and intensity of the abnormal physical features and the real-time signal quality index includes: identifying multiple simultaneous and interacting abnormal physical features; analyzing the sensitivity differences of the measurement principle of each sensing element to different combinations of abnormal physical features; establishing a multi-dimensional influence factor matrix based on the interacting abnormal physical features and the sensitivity differences; dynamically calculating the comprehensive influence weight by combining the influence factor matrix and the signal quality index; and assessing the impact of the abnormal physical features on the signal reliability of each sensing element based on the comprehensive influence weight.

3. The multi-sensor liquid-liquid interface detection method according to claim 1, characterized in that, The step after adjusting the contribution weight of each sensing element in interface positioning according to the degree of reliability impact includes: calculating the fluctuation range of the signal quality index of each sensing element; adjusting the adjustment rate and adjustment step size of the contribution weight of each sensing element according to the fluctuation range of the signal quality index; increasing the monitoring frequency of the signal quality of the sensing element when the fluctuation range of the signal quality index is large; and limiting the adjustment range of the contribution weight when the signal quality index continuously exceeds the preset signal range.

4. The multi-sensor liquid-liquid interface detection method according to claim 3, characterized in that, The determination step for the signal quality index continuously exceeding the preset confidence interval includes: calculating the average value and standard deviation of the signal quality index within a preset time window; adjusting the upper and lower limits of the preset confidence interval based on the average value and the standard deviation to obtain an adjusted confidence interval; and determining that the signal quality index continuously exceeds the preset confidence interval when multiple consecutive sampling points of the signal quality index are outside the adjusted confidence interval and the deviation direction of the signal quality index is consistent.

5. The multi-sensor liquid-liquid interface detection method according to claim 4, characterized in that, The step of limiting the adjustment range of the contribution weight when the signal quality index continuously exceeds the preset confidence interval further includes: continuously monitoring multiple key process parameters when the signal quality index continuously deviates from the adjustment confidence interval for more than a preset time length; when any one or more of the key process parameters recover to the normal range within a preset time, and the signal quality index of each sensing element continuously recovers to the adjustment confidence interval within a preset time, stopping the current weight adjustment, restoring the contribution weight of each sensing element to the state before adjustment, or re-evaluating and adjusting the weight based on the recovered signal quality.

6. The multi-sensor liquid-liquid interface detection method according to claim 5, characterized in that, The step of stopping the current weight adjustment when any one or more of the key process parameters return to the normal range within a preset time, and the signal quality indicators of each sensing element continuously return to the adjustment confidence interval within the preset time, includes: when any one or more of the key process parameters return to the normal range within a preset time, and the signal quality indicators of each sensing element continuously return to the adjustment confidence interval within the preset time, the system enters an observation period; during the observation period, the system continuously monitors the stability and trend of the key process parameters; during the observation period, the system continuously monitors the stability and trend of the signal quality indicators; when the key process parameters and the signal quality indicators remain stable and without abnormal fluctuations during the observation period, the current weight adjustment is stopped.

7. The multi-sensor liquid-liquid interface detection method according to claim 5, characterized in that, The step of restoring the contribution weights of each sensing element to their pre-adjustment state or re-evaluating and adjusting the weights based on the restored signal quality includes: continuously monitoring the real-time signal quality indicators of each sensing element; determining the weight recovery curve or recovery rate of each sensing element based on the measurement principle and inherent response characteristics to the recovery state of each sensing element; asynchronously restoring the contribution weights of each sensing element based on the weight recovery curve or recovery rate of each sensing element; and restoring the weight of the sensing element to its pre-adjustment state or re-evaluating and adjusting the weights based on the restored signal quality when the signal quality indicators of the sensing element have stably recovered to the normal range.

8. A multi-sensor liquid-liquid interface detection system, characterized in that, The system includes: a raw data acquisition module for acquiring multi-source, multi-height interface physical property data and preprocessing the data to obtain sensing element signal curves; the multi-source, multi-height interface physical property data refers to physical parameter information about the liquid-liquid two-phase interface region acquired from different types of sensing elements and at different vertical heights; a model library building module for building a response model library for multiple sensors to different interface states, the response model library containing response curves of typical signals of each sensing element under different interface transition layer states and corresponding interface state descriptions; a real-time data acquisition module for acquiring the response curve of each sensing element in real time; and a weight adjustment module for converting the real-time response curve of each sensing element into a multi-dimensional real-time feature vector, the real-time feature vector containing the curve shape descriptor, gradient features, local fluctuation features, and curve complexity index, and calculating the relationship between each real-time feature vector and the response model. The Euclidean distance between feature vectors of all known interface states in the model library is used to calculate an unknown score. When the unknown score exceeds a preset threshold, the current interface state is marked as an abnormal unknown state. When the current interface state is marked as an abnormal unknown state, abnormal physical features are identified and quantified based on the specific features that contribute the most to high deviation in the real-time feature vectors. The real-time signal quality indicators of each sensing element are monitored, including the signal-to-noise ratio, signal stability, and the degree of deviation between the signal and background noise. Based on the type and intensity of the abnormal physical features and the real-time signal quality indicators, the impact of the abnormal physical features on the signal reliability of each sensing element is evaluated. Based on the degree of reliability impact, the contribution weight of each sensing element in interface positioning is adjusted. The positioning output module is used to determine the core position or feature region of the interface based on the adjusted contribution weight and the preset interface positioning rules.

Images