Filling flow detection method and system for filling production line

Abstract

The invention relates to the technical field of electric digital data processing, in particular to a filling flow detection method and system for a filling production line, and the method comprises the steps: obtaining filling flow data at each moment in the current filling process; and inputting the filling flow data at each moment in the current filling process into an improved machine learning model, outputting the liquid amount of the inner wall of the pipeline after the current filling is stopped, and if the liquid amount of the inner wall of the pipeline after the current filling is stopped is greater than a set threshold value, judging that the current filling is abnormal. The problem that existing model detection precision is not high is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electric digital data processing. More particularly, the present application relates to a filling flow detection method and system for a filling production line. BACKGROUND

[0002] Currently, in the fields of beverage, condiment, chemical liquid, etc., real-time detection and abnormal identification of filling flow on the filling production line are key links to guarantee product quality, reduce raw material waste, and maintain production stability. Traditional filling flow detection methods mostly rely on hardware devices such as flow meters and pressure sensors to collect flow data, and combine fixed threshold judgment and simple statistical analysis to determine whether there is an abnormality in the filling process. Although this method is relatively simple to implement, it has several shortcomings. First, the fixed threshold judgment method has poor adaptability to different batches and different working conditions. When the viscosity, temperature or filling pressure of the liquid changes, the threshold often needs to be manually reset, otherwise false positives or false negatives may occur. Second, simple statistical analysis methods cannot effectively handle the inherent random noise in the collected data. Especially when the production line is running at high speed, a large number of transient abnormalities caused by equipment vibration, liquid fluctuation and sensor drift will be mixed in the flow signal, which will seriously affect the accuracy of the judgment result.

[0003] In recent years, with the rapid development of intelligent manufacturing and industrial big data technology, machine learning methods have been increasingly applied to the field of flow detection in the filling process. Through deep analysis and modeling of a large amount of historical flow data accumulated on the production line, machine learning models can automatically mine the potential nonlinear mapping relationship between flow changes and abnormal states, thereby to some extent, getting rid of the dependence on artificial experience and fixed thresholds of traditional methods, and realizing more adaptive and predictive abnormal detection. Compared with simply relying on artificial rules or simple statistical methods, this kind of data-driven model can not only consider multiple features such as instantaneous value, fluctuation amplitude, trend change of flow, but also can continuously improve the recognition ability of complex working conditions through continuous training and optimization, especially in the face of multi-batch, multi-variety and variable equipment state filling production environment, showing higher flexibility and intelligent level.

[0004] However, the existing machine learning-based models still have some limitations that cannot be ignored. First, most models have fixed values for kernel function parameters (such as length scale) or other hyperparameters, lacking adaptive adjustment ability for dynamic changes in the filling process, which leads to the problem of low detection accuracy when the model faces filling processes with different stability levels, which is too sensitive to noise or not sensitive enough to subtle abnormalities. SUMMARY

[0005] To address the problem of low detection accuracy mentioned in the background art, the present invention provides solutions in the following aspects.

[0006] In a first aspect, the present invention provides a filling flow detection method for a filling production line, comprising: acquiring filling flow data at each moment during the current filling process; inputting the filling flow data at each moment during the current filling process into an improved machine learning model, outputting the liquid volume on the inner wall of the pipe after the current filling stops; if the liquid volume on the inner wall of the pipe after the current filling stops is greater than a set threshold, then determining that there is an anomaly in the current filling; wherein, the improved machine learning model includes a length scale in a Gaussian kernel function, the length scale being positively correlated with the initial value and negatively correlated with the update factor of the current filling, the update factor representing the stability of the current filling process.

[0007] The aforementioned technical solution inputs the flow data from the filling process into an improved machine learning model with adaptive adjustment capabilities. During the analysis, it dynamically balances the model's sensitivity to abnormal fluctuations with its adaptability to stable trends, thereby more accurately estimating the amount of residual liquid on the inner wall of the pipeline after filling and enabling rapid anomaly detection. By introducing a length scaling design in the Gaussian kernel function that is positively correlated with the initial value and inversely correlated with the stability index, the model can automatically adjust the smoothness of the feature space for different operating conditions. This maintains prediction accuracy in stable conditions while improving response speed and recognition capabilities when anomalies occur, effectively enhancing the reliability, adaptability, and real-time performance of filling flow detection.

[0008] Furthermore, the length scale for, In the formula, To find the maximum value function, The minimum value of the preset length scale. As the initial value, For the natural constant An exponential function with base 0. For the first The update factor for the next filling process.

[0009] The aforementioned technical solution introduces a maximum value function and an exponential decay mechanism into the calculation of the length scale, enabling the model to adaptively adjust its sensitivity to feature differences under different filling batches. When the update factor is large, the length scale decreases accordingly, thereby enhancing the model's responsiveness to abnormal changes; while when the update factor is small or close to normal, the length scale remains above a preset minimum to avoid overfitting and feature distortion. This design ensures both the model's prediction accuracy under stable operating conditions and improves its detection sensitivity under abnormal operating conditions, thus significantly enhancing the robustness and adaptability of filling flow detection.

[0010] Furthermore, the update factor is, , For the first The first level of abnormality in the filling process For the first The second level of abnormality in the filling process.

[0011] The above technical solution achieves a comprehensive characterization of abnormal states in the filling process by multiplying the first degree of anomaly by the second degree of anomaly to obtain the update factor. The first degree of anomaly reflects the deviation between the flow fluctuation and the historical average during a single filling process, reflecting the difference between the current operating conditions and the normal state; the second degree of anomaly describes the directionality and persistence of abnormal changes during multiple filling processes, reflecting the stable characteristics of the abnormal trend. The combination of these two factors not only captures instantaneous abnormal fluctuations but also identifies long-term abnormal trends, enabling the update factor to more comprehensively and accurately characterize the stability of the filling process. This provides a more reliable basis for subsequent adaptive adjustment of model parameters and precise anomaly detection.

[0012] Furthermore, the first degree of anomaly for, , For the first The first level of abnormality in the filling process To preset hyperparameters, For the first Dynamic flow resistance during the filling process This is the average of the dynamic flow resistance over multiple historical filling processes.

[0013] The above technical solution calculates the degree of anomaly by using a weighted recursive update method, combining the relative deviation between the dynamic flow resistance of the current filling process and the historical average level with the degree of the previous anomaly, effectively balancing the response speed and stability of anomaly detection.

[0014] Furthermore, the dynamic flow resistance for, , For the first Flow variance of each filling process For the first Average flow rate per filling process To preset hyperparameters, For the natural constant An exponential function with base 0. For the first Flow deviation during the filling process.

[0015] The technical solution comprehensively constructs a dynamic flow resistance index by three key statistical characteristics of variance, average value and skewness of flow, uses the ratio of variance and average flow to reflect the relative intensity of flow fluctuation, and effectively captures the instability and abnormal fluctuation characteristics of flow in the filling process by amplifying the influence of asymmetry of flow distribution through an exponential function. This multi-dimensional feature fusion method not only improves the sensitivity to the flow fluctuation law, but also enhances the identification ability to abnormal state, so that the dynamic flow resistance can more accurately reflect the actual stability of the filling process, thereby providing more scientific and comprehensive data support for abnormal detection and model parameter adjustment.

[0016] Further, the second abnormality degree , , is a symbol function, , is the first abnormality degree of the first filling process, is the first abnormality degree of the second filling process, is the first abnormality degree of the nth filling process. is the total number of historical multiple filling processes.

[0017] The technical solution comprehensively constructs a dynamic flow resistance index by three key statistical characteristics of variance, average value and skewness of flow, uses the ratio of variance and average flow to reflect the relative intensity of flow fluctuation, and effectively captures the instability and abnormal fluctuation characteristics of flow in the filling process by amplifying the influence of asymmetry of flow distribution through an exponential function. This multi-dimensional feature fusion method not only improves the sensitivity to the flow fluctuation law, but also enhances the identification ability to abnormal state, so that the dynamic flow resistance can more accurately reflect the actual stability of the filling process, thereby providing more scientific and comprehensive data support for abnormal detection and model parameter adjustment.

[0018] Further, the method further includes denoising the filling flow data.

[0019] Further, the machine learning model is a support vector machine model.

[0020] Further, the method further includes training the machine learning model, and specifically, inputting a training set into a pre-constructed machine learning model for training, calculating a loss between a predicted value and a label during the training process, adjusting model parameters by a gradient descent method to minimize the prediction error, iteratively adjusting the parameters of the machine learning model until the loss is less than a set value or a set number of training times is reached, and finally obtaining the trained machine learning model.

[0021] In a second aspect, the present application provides a filling flow detection system for a filling production line, comprising a memory and a processor, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the computer program instructions realize the filling flow detection method for the filling production line according to any one of the preceding aspects.

[0022] The present application has the advantages of: The present application intelligently analyzes the flow data at each moment in the filling process by combining an improved machine learning model, realizes accurate prediction of the amount of residual liquid on the inner wall of the pipeline after filling is completed, and effectively determines filling abnormalities based on this. The length scale adaptive adjustment mechanism used can dynamically optimize the perception range of the model according to the stability of the filling process, improving the ability to capture flow changes under different working conditions, thereby enhancing the sensitivity and accuracy of abnormal detection. The multi-dimensional characterization of the abnormal degree and the update factor formed by its fusion can not only comprehensively reflect instantaneous fluctuations and long-term trends, but also effectively suppress noise interference, improving the robustness of the model. Through denoising processing of the flow data and training and optimization of the support vector machine model, this scheme significantly improves the ability to identify abnormal states in complex filling environments and real-time response capabilities, realizes efficient, stable and adaptive filling flow detection, and greatly promotes the level of automatic monitoring of the production line and product quality assurance. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 is a flow chart schematically showing a filling flow detection method for a filling production line according to an embodiment of the present application; Figure 2 is a structural block diagram schematically showing a filling flow detection system for a filling production line according to an embodiment of the present application. DETAILED DESCRIPTION

[0024] A filling flow detection method for a filling production line embodiment.

[0025] As Figure 1 shown, the flow chart of a filling flow detection method for a filling production line according to an embodiment of the present application includes the following steps: S1: Obtain the filling flow data at each moment in the current filling process.

[0026] In a preferred embodiment, in order to accurately obtain the filling flow data at each moment in the current filling process, the present embodiment is provided with a high-precision flow sensor on the liquid delivery pipeline of the filling production line. The sensor is preferably an electromagnetic flowmeter or a mass flowmeter to ensure that it can still maintain high measurement accuracy under different liquid physical properties (such as viscosity, density, temperature).

[0027] In practical applications, when the filling device starts to perform a filling task of a batch, a start instruction is first sent to the flow sensor to make it enter a real-time monitoring mode. At this time, the flow sensor continuously collects the instantaneous flow of the liquid in the pipeline at a preset sampling frequency, and each sampling period generates a flow data point with a time stamp information accurate to milliseconds to ensure that the complete flow change curve can be restored in the subsequent analysis process.

[0028] Further, the filling flow data is denoised. In order to further improve the accuracy and availability of the filling flow data obtained at each moment in the current filling process, the filling flow data is denoised after data collection. The denoising can use mean filtering to eliminate high-frequency noise and abnormal fluctuations introduced by measurement error, environmental interference and equipment mechanical fluctuation, so as to obtain a more smooth and stable flow curve that truly reflects the liquid flow state.

[0029] S2: inputting the filling flow data at each moment in the current filling process into the improved machine learning model to output the amount of liquid on the inner wall of the pipeline after the current filling stops.

[0030] In a preferred embodiment, the machine learning model is a support vector machine model. Further, the support vector machine model is trained as follows. The training set is input into the pre-constructed support vector machine model for training. During the training process, the loss between the predicted value and the label is calculated. The model parameters are adjusted by gradient descent method to minimize the prediction error. The parameters of the support vector machine model are iteratively adjusted until the loss is less than a set value or a set number of training times is reached, and finally a trained support vector machine is obtained.

[0031] The improved machine learning model includes a length scale in a Gaussian kernel function, and the length scale is , , wherein is a maximum value function, is a minimum value of the preset length scale, is an initial value, is an exponential function with a natural constant as the base, is the first The update factor for the filling process is determined by employing a dynamic adjustment strategy based on an exponential function in the length scaling design of the Gaussian kernel function, while setting a minimum threshold for the length scale to ensure it does not fall below a preset lower limit. Specifically, the length scale adaptively adjusts as the update factor changes during the filling process, flexibly shrinking or expanding according to fluctuations in flow data, thereby achieving accurate capture of data characteristics under different filling states. This mechanism effectively avoids the overfitting or underfitting problems that may arise from traditional fixed-length scaling, improving the kernel function's adaptability to local data structures.

[0032] Furthermore, by limiting the lower bound of the length scale, this scheme prevents the model from becoming sensitive to noise and the prediction results from being unstable due to an excessively small length scale, thus enhancing the overall robustness and generalization ability of the algorithm. In summary, this technical solution improves the stability and accuracy of machine learning models in filling flow data processing, enabling them to more reliably reflect the changing characteristics of the actual filling process. This provides a solid data foundation for subsequent filling volume prediction and anomaly detection, significantly optimizing the control effect and production efficiency of the filling process.

[0033] The update factor is, , For the first The first level of abnormality in the filling process For the first The second level of abnormality in the filling process.

[0034] First degree of abnormality for, , For the first The first level of abnormality in the filling process To preset hyperparameters, For the first Dynamic flow resistance during the filling process This is the average dynamic flow resistance over multiple historical filling processes. The dynamic flow resistance... for, , For the first Flow variance in each filling process For the first Average flow rate per filling process To preset hyperparameters, For the natural constant An exponential function with base 0. For the first Flow deviation during the filling process.

[0035] By constructing an update factor composed of two types of anomaly levels, and incorporating a combination of historical weight smoothing and current flow resistance deviation in the calculation of the first anomaly level, the system can balance sensitivity to short-term fluctuations in the filling process with stability to long-term trends. The calculation of dynamic flow resistance integrates multi-dimensional features such as flow variance, average flow rate, and flow distribution skewness, and utilizes an exponential function to amplify asymmetric features, thus providing a more comprehensive reflection of liquid flow stability and anomaly levels. This approach not only captures subtle anomalies in the filling process in a timely manner but also suppresses the interference of occasional fluctuations on the overall assessment, helping to improve the model's discrimination accuracy and robustness under complex operating conditions. It provides high-quality input for subsequent model parameter adjustments based on the update factor, thereby significantly optimizing the prediction and control performance of the filling process.

[0036] Second degree of abnormality , , For symbolic functions, , The first sequence The first level of abnormality in the filling process This represents the total number of filling processes throughout history.

[0037] By introducing a sign function into the calculation of the second degree of anomaly to statistically analyze the changing direction of the first degree of anomaly in the continuous filling process, and averaging the results over multiple historical data points, this method effectively reflects the persistence and trend of the anomaly degree over time. Compared to assessment methods that rely solely on the magnitude of a single anomaly, this method can identify the frequency and directional consistency of anomaly changes, thus distinguishing between occasional fluctuations under a stable trend and truly persistent, evolving anomalies. This approach provides a more robust ability to identify anomaly trends in dynamic monitoring, reduces interference from short-term random noise, provides a more accurate trend reference for the predictive control of the filling process, and enhances stability and early warning capabilities under complex operating conditions.

[0038] S3: If the amount of liquid on the inner wall of the pipe is greater than the set threshold after the current filling stops, it is determined that there is an abnormality in the current filling.

[0039] In a preferred embodiment, the direct determination of filling process abnormalities is achieved by detecting the amount of residual liquid on the inner wall of the pipeline after filling is completed and comparing it with a preset threshold. Since the residual liquid in the pipeline should normally be maintained at a stable and low level after filling is stopped, exceeding the threshold often means that there is an abnormality in the filling pressure, flow control or liquid recovery, etc. This method can quickly locate potential problems without relying on complex signal analysis. This physical residual-based judgment mechanism has the advantages of strong intuitiveness, high reliability and easy implementation, and can achieve timely abnormal identification in actual production, reduce liquid waste and production efficiency due to equipment failure or control deviation, and provide clear basis for subsequent maintenance and parameter optimization.

[0040] The scheme of the present application realizes accurate analysis and abnormality determination of filling flow data by introducing a machine learning model based on improved support vector machines, combining dynamic adjustment of Gaussian kernel length scale and multi-dimensional abnormality degree index. Through denoising processing of flow data and training optimization of the model, the scheme can effectively improve the identification accuracy and real-time response capability of filling process abnormalities, enhance the stability and robustness of detection, and further improve the quality control level and production efficiency of the filling production line.

[0041] An embodiment of a filling flow detection system for a filling production line: As shown in Figure 2 The structure block diagram of an embodiment of a filling flow detection system for a filling production line of the present application includes a processor and a memory.

[0042] The present application also provides a filling flow detection system for a filling production line. As shown in Figure 2 The system includes a processor and a memory, and the memory stores computer program instructions which, when executed by the processor, implement the filling flow detection method for a filling production line according to the present application.

[0043] The filling flow detection system for a filling production line also includes a communication interface and other components known to those skilled in the art, the settings and functions of which are known in the art, and therefore will not be described here.

[0044] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device. Any application or module described in this invention can be implemented using computer-readable / executable instructions stored or otherwise maintained by such a computer-readable medium.

[0045] In the description of this specification, "multiple" or "several" means at least two, such as two, three or more, unless otherwise explicitly specified.

[0046] While this specification has shown and described numerous embodiments of the invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention. It should be understood that various alternatives to the embodiments of the invention described herein may be employed in the practice of this invention.

Claims

1. A method for detecting filling flow rate in a filling production line, characterized in that, include: Obtain filling flow data at each moment during the current filling process; The filling flow rate data at each moment during the current filling process is input into the improved machine learning model, and the liquid volume on the inner wall of the pipe after the current filling stops is output. If the liquid volume on the inner wall of the pipe after the current filling stops is greater than a set threshold, it is determined that there is an abnormality in the current filling. The improved machine learning model includes a length scale in a Gaussian kernel function, which is positively correlated with the initial value and negatively correlated with the update factor of the current filling process, wherein the update factor characterizes the stability of the current filling process.

2. The filling flow detection method for a filling production line according to claim 1, characterized in that, The length scale for, In the formula, To find the maximum value function, The minimum value of the preset length scale. As the initial value, For the natural constant An exponential function with base 0. For the first The update factor for the next filling process.

3. The filling flow detection method for a filling production line according to claim 1, characterized in that, The update factor is, , For the first The first level of abnormality in the filling process For the first The second level of abnormality in the filling process.

4. The filling flow detection method for a filling production line according to claim 3, characterized in that, First degree of abnormality for, , For the first The first level of abnormality in the filling process To preset hyperparameters, For the first Dynamic flow resistance during the filling process This is the average of the dynamic flow resistance over multiple historical filling processes.

5. The filling flow detection method for a filling production line according to claim 4, characterized in that, The dynamic flow resistance for, , For the first Flow variance of each filling process For the first Average flow rate per filling process To preset hyperparameters, For the natural constant An exponential function with base 0. For the first Flow deviation during the filling process.

6. The filling flow detection method for a filling production line according to claim 3, characterized in that, Second degree of abnormality , , For symbolic functions, , The first sequence The first level of abnormality in the filling process This represents the total number of filling processes throughout history.

7. The filling flow detection method for a filling production line according to claim 1, characterized in that, It also includes denoising the filling flow data.

8. The filling flow detection method for a filling production line according to claim 1, characterized in that, The machine learning model is a support vector machine model.

9. A method for detecting filling flow rate in a filling production line according to claim 1, characterized in that, This also includes training machine learning models, specifically: The training set is fed into a pre-built machine learning model for training. During the training process, the loss between the output prediction value and the label is calculated. Use gradient descent to adjust model parameters to minimize prediction error; The parameters of the machine learning model are iteratively adjusted until the loss is less than a set value or the set number of training iterations are reached, and finally a well-trained machine learning model is obtained.

10. A filling flow detection system for a filling production line, characterized in that, It includes a memory and a processor, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, it implements the filling flow detection method for a filling production line as described in any one of claims 1 to 9.

Images