Fusion type filter element residual life prediction device and method based on KAN network structure

Abstract

The invention is suitable for the technical field of filter element service life prediction, and particularly relates to a fusion type filter element residual service life prediction device and method based on a KAN network structure, in the device, a power adapter is connected with a power conversion circuit and a booster pump, the power conversion circuit is connected with a sensor group and an embedded module, the booster pump is connected with the embedded module, and the sensor group is connected with the embedded module. The embedded module is used for collecting detection data from the sensor group and the booster pump and transmitting the detection data to the cloud platform. According to the method, factors influencing the service life of the filter element, such as the rotating speed of the booster pump, the TDS before and after the membrane and the flow, are fully considered, a prediction algorithm based on a filter element purification mechanism model and a data driving prediction algorithm of a KAN network structure are fused, the effect of accurately predicting the residual service life of the filter element in real time can be achieved, on one hand, the usability of the filter element is monitored in real time, and on the other hand, the service life of the filter element is prolonged; the drinking safety is ensured; on the other hand, the filter element is replaced before being scrapped, the value of the filter element is utilized to the maximum extent, and resource waste is avoided.

Description

Technical Field

[0001] The present invention belongs to the technical field of filter element life prediction, and in particular relates to a fusion filter element remaining life prediction device and method based on a KAN network structure. Background Art

[0002] As the core component of the water purifier, the filter element is crucial to the effective purification of the entire device. Limited by the filtering mechanism, as the number of trapped pollutants continues to increase, the filtering performance of the filter element will inevitably decay, and it often needs to be replaced after being blocked to a certain extent. Accurately predicting the remaining useful life (RUL) of the filter element under different usage conditions can not only ensure the safe and reliable operation of the system, but also maximize the use of the residual value of the filter element.

[0003] The existing filter element RUL usually adopts a simple countdown, setting a target life value when it is put into use. When the cumulative use time reaches the target life value, the filter element is judged to have reached its life. This method is simple and crude. When the incoming water quality is poor, the filter element may be scrapped prematurely, resulting in the water purifier being unable to purify; when the incoming water quality is good, the filter element may continue to be used normally, and premature replacement will lead to unnecessary waste of resources. Summary of the invention

[0004] The purpose of the present invention is to provide a fusion filter element remaining life prediction device based on a KAN network structure, aiming to solve the problem that the existing prediction methods cannot achieve accurate prediction and easily lead to waste of resources.

[0005] The present invention is implemented as follows: a fusion filter element remaining life prediction device based on a KAN network structure, the device comprising a power adapter, a power conversion circuit, a booster pump, a sensor group and an embedded module, the power adapter is connected to the power conversion circuit and the booster pump, the power conversion circuit is connected to the sensor group and the embedded module, the booster pump is connected to the embedded module, and the embedded module is used to collect detection data from the sensor group and the booster pump and transmit it to a cloud platform;

[0006] The embedded module regularly transmits the detection data to the cloud platform. After analyzing all the detection data through the RUL prediction algorithm on the cloud platform, the latest RUL prediction value is obtained. When the latest RUL prediction value reaches the set target value, the embedded module drives the buzzer through the PWM signal to sound an alarm, prompting the user to replace the filter element.

[0007] Preferably, the sensor group includes a pre-membrane TDS sensor, a post-membrane TDS sensor, a flow sensor and a buzzer.

[0008] Another object of the present invention is to provide a method for predicting the remaining life of a fused filter element based on a KAN network structure, which is applied to the above-mentioned fused filter element remaining life prediction device based on a KAN network structure, and the method comprises:

[0009] Acquire detection data through a fusion filter element remaining life prediction device based on a KAN network structure, and store the detection data, wherein the detection data includes booster pump speed, TDS before the membrane, TDS after the membrane, and flow data;

[0010] Different life prediction models are selected according to the working status of the direct drinking machine. The life prediction models include prediction models based on the filter element purification mechanism and prediction models based on data drive;

[0011] The life prediction model is converted into the form of a state space equation, and the historical data is filtered through a filtering algorithm to obtain filtered data. When in the water production state, the historical data is the life curve up to the current time, and when in the standby state, the historical data is the detection data before the current time;

[0012] The life prediction model in the form of state space equation is used to predict the life and obtain the first life prediction value RUL KAN ;

[0013] The life prediction model based on the filter element purification mechanism is used to predict the life and obtain the second life prediction value RUL 模型 ;

[0014] The first predicted life value and the second predicted life value are weighted to obtain the fused predicted life value RUL 最终值 , prompting the user based on the fused predicted life span value.

[0015] Preferably, the first life prediction value and the second life prediction value are weighted to obtain a fused predicted life value RUL 最终值 In the step of

[0016] RUL 最终值 =0.3×RUL 模型 +0.7×RUL KAN .

[0017] Preferably, in the step of converting the life prediction model into a state space equation, the state space equation is expressed as:

[0018] State parameter X i =[a k ; b k ;…;n k ]

[0019] Equation of state X i =f(Xk-1 )+W k

[0020] Output equation C k =h(X k ,T k )+V k

[0021] Among them, a k , b k ,…,n k Where a, b, ..., n represent the various factors that constitute the state parameter matrix X, the subscript k is the cumulative number of iterations, and k-1 is the previous iteration of k;

[0022] f(X k-1 ) is based on X k-1 and X k The function established by the parameter relationship of

[0023] W k is a constant that affects the state equation;

[0024] T k is the cumulative usage time when the cumulative iteration number is k;

[0025] C k T k Availability at a given point in time;

[0026] h(X k ,T k ) is based on X k and T k The function of establishing the relationship;

[0027] V k is a constant that affects the output equation.

[0028] Preferably, in standby mode, the calculation formula for the filter element life is:

[0029]

[0030] Among them, m max The theoretical maximum adsorption capacity of the filter element is determined by laboratory measurement data of the same type of filter element;

[0031] ρ is the concentration of particle pollutants, which is converted from the TDS value before the membrane;

[0032] Q is the maximum water output, that is, the maximum flow rate detected by the flow sensor during the water production process;

[0033] μ is the filtration efficiency, which depends on the filter element and is a fixed value;

[0034] t is the running time, which is obtained by counting the cumulative water production time on the cloud platform;

[0035] v is the pump speed at the current moment.

[0036] Preferably, in the water production state, the remaining life curve of the filter element is expressed as:

[0037] K=(φ L-1 ·φ L-2 ·…φ1·φ0)·X

[0038] in, is the function matrix corresponding to the i-th KAN layer, X is the input matrix, and the data types included are booster pump speed, post-membrane TDS value, and flow value.

[0039] Preferably, the filtering algorithm includes a Kalman filtering algorithm and a particle filtering algorithm. When the historical data is processed, it is first processed by the Kalman filtering algorithm and then processed by the particle filtering algorithm.

[0040] Preferably, in the step of prompting the user based on the fused predicted life value, the fused predicted life value is compared with a preset life value, and prompts of different intensities are given to the user.

[0041] The integrated filter element remaining life prediction device based on the KAN network structure provided by the present invention fully considers the factors affecting the filter element life, including the boost pump speed, TDS before and after the membrane, and flow rate, and integrates the prediction algorithm based on the filter element purification mechanism model and the data-driven prediction algorithm of the KAN network structure. It can accurately predict the remaining life of the filter element in real time. On the one hand, it monitors the availability of the filter element in real time to ensure drinking water safety; on the other hand, it replaces the filter element when it is close to being scrapped, thereby maximizing the value of the filter element and avoiding waste of resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 An architecture diagram of a fusion filter element remaining life prediction device based on a KAN network structure provided in an embodiment of the present invention;

[0043] Figure 2 An overall flow chart of the fusion filter element remaining life prediction method based on the KAN network structure provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0045] It is understood that the terms "first", "second", etc. used in this application may be used herein to describe various elements, but unless otherwise specified, these elements are not limited by these terms. These terms are only used to distinguish a first element from another element. For example, without departing from the scope of this application, a first xx script may be referred to as a second xx script, and similarly, a second xx script may be referred to as a first xx script.

[0046] like Figure 1 , which is an architecture diagram of a fusion filter element remaining life prediction device based on a KAN network structure provided in an embodiment of the present invention, wherein the device comprises a power adapter, a power conversion circuit, a booster pump, a sensor group and an embedded module, wherein the power adapter is connected to the power conversion circuit and the booster pump, the power conversion circuit is connected to the sensor group and the embedded module, the booster pump is connected to the embedded module, and the embedded module is used to collect detection data from the sensor group and the booster pump and transmit it to a cloud platform;

[0047] The embedded module regularly transmits the detection data to the cloud platform. After analyzing all the detection data through the RUL prediction algorithm on the cloud platform, the latest RUL prediction value is obtained. When the latest RUL prediction value reaches the set target value, the embedded module drives the buzzer through the PWM signal to sound an alarm, prompting the user to replace the filter element.

[0048] The sensor group includes a pre-membrane TDS sensor, a post-membrane TDS sensor, a flow sensor and a buzzer

[0049] In this embodiment, after the direct drinking machine takes power from the municipal electricity 220V socket, it is stepped down by the power adapter and the power conversion circuit to provide power for each working part, wherein the working voltage of the booster pump is 24V, the working voltage of the pre-membrane TDS sensor, the post-membrane TDS sensor, the flow sensor and the buzzer is 5V, and the working voltage of the embedded module 03 is 3.3V. After each working part is powered on, it is in a dormant state. When there is a water intake signal to drive the whole machine to start water production, the booster pump starts. At the same time, the pre-membrane TDS sensor, the post-membrane TDS sensor and the flow sensor start working, and the TDS values ​​of the incoming water and the filtered water and the flow in the filtered water pipeline are monitored in real time. The embedded module obtains the speed data of the booster pump through the FG feedback signal, and obtains the pre-membrane TDS, post-membrane TDS and flow values ​​through the communication interface. As time goes by, the water production volume continues to increase, and the filter element is gradually blocked by the intercepted pollutants, resulting in an increase in the flow resistance, which makes the workload of the booster pump continue to increase. At this time, the speed of the booster pump will decrease, the post-membrane TDS will increase, and the flow rate will decrease.

[0050] The embedded module transmits the booster pump speed, TDS before the membrane, TDS after the membrane and flow data to the cloud platform regularly via 4G. After analyzing all the data through the RUL prediction algorithm on the cloud platform, the latest RUL prediction value (i.e. RUL 最终值 ). When the latest RUL prediction value reaches the set target value, the embedded module drives the buzzer to sound an alarm through the PWM signal, prompting the user to replace the filter element.

[0051] like Figure 2 As shown, the fusion filter element remaining life prediction method based on the KAN network structure provided by the embodiment of the present invention is applied to the fusion filter element remaining life prediction device based on the KAN network structure as described above, and the method includes:

[0052] The detection data is acquired through a fusion filter element remaining life prediction device based on a KAN network structure, and the detection data is stored. The detection data includes booster pump speed, TDS before the membrane, TDS after the membrane and flow data.

[0053] In this embodiment, the booster pump speed, TDS before the membrane, TDS after the membrane and flow rate data transmitted back by the embedded module are received and stored in the cloud database.

[0054] Different life prediction models are selected according to the working status of the direct drinking machine. The life prediction models include prediction models based on the filter element purification mechanism and prediction models based on data-driven.

[0055] In this embodiment, when the direct drinking machine is in standby mode, the computing resources are relatively abundant. The prediction model based on the filter element purification mechanism can be used to evaluate the health status of the filter element in detail and accurately predict the remaining life span RUL of the filter element. 模型 .

[0056]

[0057] Among them, m max The theoretical maximum adsorption capacity of the filter element is determined by laboratory measurement data of the same type of filter element;

[0058] ρ is the concentration of particle pollutants, which is converted from the TDS value before the membrane;

[0059] Q is the maximum water output, that is, the maximum flow rate detected by the flow sensor during the water production process;

[0060] μ is the filtration efficiency, which depends on the filter element and is a fixed value;

[0061] t is the running time, which is obtained by counting the cumulative water production time on the cloud platform;

[0062] v is the pump speed at the current moment.

[0063] When the direct drinking machine is in the water-making state, the data-driven prediction model can directly use historical data (life curve) to establish a machine learning model to predict the future aging trend of the filter element without relying on a specific physical model, which can better realize the online application in the filter element life management system, and obtain RULKAN through algorithm extrapolation. The KAN network structure is adopted in the present invention, combining the advantages of multi-layer perceptron (MLP) and spline curve (Spline).

[0064] The remaining life curve of the filter element is expressed as:

[0065] K=(φ L-1 ·φ L-2 ·…φ1·φ0)·X

[0066] in, is the function matrix corresponding to the i-th KAN layer, X is the input matrix, and the data types included are booster pump speed, post-membrane TDS value, and flow value.

[0067] The life prediction model is converted into the form of a state space equation, and the historical data is filtered through a filtering algorithm to obtain filtered data. When in the water production state, the historical data is the life curve up to the current time. When in the standby state, the historical data is the detection data before the current time.

[0068] In this embodiment, two filtering algorithms, Kalman filtering (KF) and particle filtering (PF), are used. Among them, the KF algorithm is an optimized autoregressive data processing algorithm that effectively suppresses Gaussian noise. It is better for filtering system processes and can effectively filter out sudden changes in signals. The PF algorithm approximates the probability density function by finding a group of random samples propagating in the state space, and replaces the integral operation with the sample mean, which has a better inhibitory effect on nonlinear Gaussian noise. For historical data, the KF algorithm is processed first, and then the PF algorithm is processed.

[0069] In the step of converting the life prediction model into the form of state space equation, the state space equation is expressed as:

[0070] State parameter X i =[a k ; b k ;…;n k ]

[0071] Equation of state X i =f(X k-1 )+W k

[0072] Output equation C k =h(X k ,Tk )+V k

[0073] Among them, a k , b k ,…,n k Where a, b, ..., n represent the various factors that constitute the state parameter matrix X, the subscript k is the cumulative number of iterations, and k-1 is the previous iteration of k;

[0074] f(X k-1 ) is based on X k-1 and X k The function established by the parameter relationship of

[0075] W k is a constant that affects the state equation;

[0076] T k is the cumulative usage time when the cumulative iteration number is k;

[0077] C k T k Availability at a given point in time;

[0078] h(X k ,T k ) is based on X k and T k The function of establishing the relationship;

[0079] V k is a constant that affects the output equation.

[0080] And initialize the filtering algorithm and state space equations.

[0081] The life prediction model in the form of state space equation is used to predict the life and obtain the first life prediction value RUL KAN ;

[0082] The life prediction model based on the filter element purification mechanism is used to predict the life and obtain the second life prediction value RUL 模型 ;

[0083] The first predicted life value and the second predicted life value are weighted to obtain the fused predicted life value RUL 最终值 , prompting the user based on the fused predicted life span value.

[0084] In this embodiment, based on the historical data before the prediction starting point, the model parameters are continuously updated using a filtering algorithm until the prediction starting point, and finally a fitting curve is formed with the horizontal axis being the accumulated usage time and the vertical axis being the filter element availability.

[0085] The remaining life of the filter element is predicted by extrapolating the model with updated parameters at the prediction starting point.

[0086] During this process, if the direct drinking machine is in standby state, the state parameters are constructed using a prediction model based on the filter purification mechanism; if the direct drinking machine is in water making state, the state parameters are constructed based on a data-driven prediction model.

[0087] The extrapolation step is 1 day, and the extrapolation simulation stops when the availability falls below the set fault threshold, which is usually set to 10% of the total usable life of the filter element. The prediction algorithm of the KAN network structure has requirements for the amount of historical data, and needs to be continuously updated through the filtering algorithm to track the nonlinear aging behavior of the filter element. To this end, when the filter element has just been put into use and the RUL prediction time scale is long, a model-based prediction algorithm is used to construct a "virtual observation value" at a future moment, so as to continuously correct the deviation during the initial extrapolation value fitting process of the KAN network structure, thereby improving the accuracy of the early RUL prediction.

[0088] When the direct drinking machine is in standby mode, the RUL is obtained by extrapolating the above steps. 模型 ; When the direct drinking machine is in the water-making state, the RUL is obtained by extrapolating the above steps KAN The final RUL is calculated according to the following weights: 最终值 , and displayed on the direct drinking machine control panel.

[0089] RUL 最终值 =0.3×RUL 模型 +0.7×RUL KAN

[0090] RUL 最终值 ≤7 days, when the user starts to take water, the buzzer will emit a slow beep to remind the user to change the cartridge;

[0091] RUL 最终值 ≤3 days, when the user starts to take water, the buzzer will emit a rapid beep to remind the user to change the cartridge.

[0092] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

[0093] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A fusion filter element remaining life prediction device based on KAN network structure, characterized in that: The device includes a power adapter, a power conversion circuit, a booster pump, a sensor group and an embedded module, the power adapter is connected to the power conversion circuit and the booster pump, the power conversion circuit is connected to the sensor group and the embedded module, the booster pump is connected to the embedded module, and the embedded module is used to collect detection data from the sensor group and the booster pump and transmit it to the cloud platform; The embedded module regularly transmits the detection data to the cloud platform. After analyzing all the detection data through the RUL prediction algorithm on the cloud platform, the latest RUL prediction value is obtained. When the latest RUL prediction value reaches the set target value, the embedded module drives the buzzer through the PWM signal to sound an alarm, prompting the user to replace the filter element.

2. The fusion filter element remaining life prediction device based on KAN network structure according to claim 1 is characterized in that: The sensor group includes a pre-membrane TDS sensor, a post-membrane TDS sensor, a flow sensor and a buzzer.

3. A fusion filter element remaining life prediction method based on KAN network structure is characterized by: Applied to the fusion filter element remaining life prediction device based on the KAN network structure as claimed in claim 1 or 2, the method comprises: Acquire detection data through a fusion filter element remaining life prediction device based on a KAN network structure, and store the detection data, wherein the detection data includes booster pump speed, TDS before the membrane, TDS after the membrane, and flow data; Different life prediction models are selected according to the working status of the direct drinking machine. The life prediction models include prediction models based on the filter element purification mechanism and prediction models based on data drive; The life prediction model is converted into the form of a state space equation, and the historical data is filtered through a filtering algorithm to obtain filtered data. When in the water production state, the historical data is the life curve up to the current time, and when in the standby state, the historical data is the detection data before the current time; The life prediction model in the form of state space equation is used to predict the life and obtain the first life prediction value RUL KAN ; The life prediction model based on the filter element purification mechanism is used to predict the life and obtain the second life prediction value RUL 模型 ; The first predicted life value and the second predicted life value are weighted to obtain the fused predicted life value RUL 最终值 , prompting the user based on the fused predicted life span value.

4. The method for predicting the remaining life of a fused filter element based on a KAN network structure according to claim 3 is characterized in that: The first predicted life value and the second predicted life value are weighted to obtain the fused predicted life value RUL 最终值 In the step of RUL 最终值 =0.3×RUL 模型 +0.7×RUL KAN 。 5. The method for predicting the remaining life of a fused filter element based on a KAN network structure according to claim 3 is characterized in that: In the step of converting the life prediction model into the form of state space equation, the state space equation is expressed as: State parameter X i =[a k ; b k ;…;n k ] Equation of state X i =f(X k-1 )+W k Output equation C k =h(X k ,T k )+V k Among them, a k , b k ,…,n k Where a, b, ..., n represent the various factors that constitute the state parameter matrix X, the subscript k is the cumulative number of iterations, and k-1 is the previous iteration of k; f(X k-1 ) is based on X k-1 and X k The function established by the parameter relationship of W k is a constant that affects the state equation; T k is the cumulative usage time when the cumulative iteration number is k; C k T k Availability at a given point in time; h(X k ,T k ) is based on X k and T k The function of establishing the relationship; V k is a constant that affects the output equation.

6. The method for predicting the remaining life of a fused filter element based on a KAN network structure according to claim 3 is characterized in that: In standby mode, the filter life calculation formula is: Among them, m max The theoretical maximum adsorption capacity of the filter element is determined by laboratory measurement data of the same type of filter element; ρ is the concentration of particle pollutants, which is converted from the TDS value before the membrane; Q is the maximum water output, that is, the maximum flow rate detected by the flow sensor during the water production process; μ is the filtration efficiency, which depends on the filter element and is a fixed value; t is the running time, which is obtained by counting the cumulative water production time on the cloud platform; v is the pump speed at the current moment.

7. The method for predicting the remaining life of a fused filter element based on a KAN network structure according to claim 3 is characterized in that: In the water production state, the remaining life curve of the filter element is expressed as: K=(φ L-1 ·f L-2 ·…φ1·φ0)·X in, is the function matrix corresponding to the i-th KAN layer, X is the input matrix, and the data types included are booster pump speed, post-membrane TDS value, and flow value.

8. The method for predicting the remaining life of a fused filter element based on a KAN network structure according to claim 3 is characterized in that: The filtering algorithm includes a Kalman filtering algorithm and a particle filtering algorithm. When the historical data is processed, it is first processed by the Kalman filtering algorithm and then processed by the particle filtering algorithm.

9. The method for predicting the remaining life of a fused filter element based on a KAN network structure according to claim 3 is characterized in that: In the step of prompting the user according to the fused predicted life value, the fused predicted life value is compared with the preset life value, and prompts of different intensities are given to the user.

Images