Electromagnetic fryer circuit fault detection system based on artificial intelligence

Abstract

The invention discloses an electromagnetic fryer circuit fault detection system based on artificial intelligence. The electromagnetic fryer circuit fault detection system comprises a data acquisition module, a pre-establishment module, a loss function design module, a joint optimization parameter module, a global adaptive activation function module and a fault detection module. The invention belongs to the field of circuit fault detection, and particularly relates to an electromagnetic fryer circuit fault detection system based on artificial intelligence. According to the scheme, a loss function is designed for parameter drift and peak noise caused by tiny resistance change and temperature drift, parameters are generated through bottom layer variable mapping, and all loss sensitivities are adjusted in a self-adaptive mode; detection and suppression of sensor peak noise are realized, timely detection of early and tiny faults is ensured, and false alarms are effectively avoided. A global adaptive activation function is constructed, parameter self-moderation is consistent with three types of loss sensitivity, forward activation and backward loss are amplified or suppressed in a linkage mode on the same fault feature scale, and then the fault detection effect is improved.

Description

Technical Field

[0001] The present invention relates to the field of circuit fault detection, and specifically refers to an electromagnetic deep fryer circuit fault detection system based on artificial intelligence. Background Art

[0002] An electromagnetic deep fryer circuit fault detection system is a system that can monitor parameters such as current, voltage, and temperature of the electromagnetic deep fryer circuit in real time, automatically diagnose the type and location of faults, and give early warnings in a timely manner. However, in general, the electromagnetic deep fryer circuit fault detection system has the problems that the fault criterion is not flexible enough, and once the working conditions or environment change, the missed detection or false detection rate increases; it is difficult to distinguish ordinary noise from tiny fault signals, which leads to poor accuracy of fault detection; in general, the electromagnetic deep fryer circuit fault detection system has the problem that it cannot capture the slight fluctuations caused by micro-changes in resistance, and over-responds to the spike noise of sensors, which leads to poor fault detection effect. Summary of the Invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides an electromagnetic deep fryer circuit fault detection system based on artificial intelligence. Aiming at the problems that the fault criterion of the general electromagnetic deep fryer circuit fault detection system is not flexible enough, and once the working conditions or environment change, the missed detection or false detection rate increases; it is difficult to distinguish ordinary noise from tiny fault signals, which leads to poor accuracy of fault detection, this solution constructs a healthy working condition current prediction model based on a fully connected feedforward neural network, designs loss functions for micro-resistance changes, parameter drift caused by temperature drift, and spike noise respectively, and uses underlying variable mapping to generate parameters to adaptively adjust the sensitivity of each loss; it can detect micro-changes in the inter-turn resistance of the coil within the noise level, is sensitive to medium-amplitude changes caused by temperature drift, realizes the detection and suppression of sensor spike noise, not only ensures the timely detection of early and tiny faults, but also effectively avoids false alarms; aiming at the problems that the general electromagnetic deep fryer circuit fault detection system cannot capture the slight fluctuations caused by micro-changes in resistance, and over-responds to the spike noise of sensors, which leads to poor fault detection effect, this solution constructs a global adaptive activation function on the basis of designing loss functions for micro-resistance changes, parameter drift caused by temperature drift, and spike noise; the parameter self-adaptation is consistent with the sensitivity of the three types of losses, and the forward activation and backward loss are linked to amplify or suppress on the same fault feature scale, thereby improving the fault detection effect.

[0004] The technical solution adopted by the present invention is as follows: An electromagnetic deep fryer circuit fault detection system based on artificial intelligence provided by the present invention includes a data acquisition module, a pre-establishment module, a loss function design module, a joint optimization parameter module, a global adaptive activation function module, and a fault detection module;

[0005] The data acquisition module acquires the characteristic data of the electromagnetic deep fryer in a healthy state;

[0006] The pre - established module constructs a feed - forward neural network model based on the standardized feature vectors in the healthy state;

[0007] The loss function design module introduces three loss functions, which are respectively for tiny resistance changes, medium fluctuations, and spike noises;

[0008] The joint optimization parameter module generates the parameters of the loss term through a learnable fractional - order exponential mapping and automatically adjusts them together with the network parameters during training through backpropagation;

[0009] The global adaptive activation function module calculates the global activation parameter based on the fractional - order exponent of the three losses and constructs an adaptive activation function;

[0010] The fault detection module performs fault detection on the newly input data.

[0011] Furthermore, the data acquisition module selects key nodes on the power supply side and at both ends of the inductance coil from the electromagnetic fryer to collect the voltage and current time - series signals in the healthy state; performs short - time Fourier transform on the time series to extract the frequency - domain and time - domain feature vectors; and performs zero - mean and unit - variance standardization on the feature vectors of all dimensions.

[0012] Furthermore, the pre - established module is to establish a model in the healthy state, with the standardized feature vectors as the input, and constructs a healthy - condition current prediction model based on the fully - connected feed - forward neural network, expressed as: ; The residual is expressed as: ; where, is the predicted value of the current in the k - th frame; is the actual value; F(·) is the fully - connected feed - forward neural network mapping, are the network parameters.

[0013] Furthermore, the loss function design module is to construct a fractional - order loss function; utilize the short - memory property of the fractional - order loss to characterize the residual, and divide the residual - driven loss into three terms, corresponding to different fault modes respectively; for the resistance change of the tiny coil, define the residual - sensitive detection loss , expressed as: ; ; where, x is the input of the loss function, taking the residual ; c is the scaling constant; is the fractional - order exponent; is the gamma function; is the gradient; is the sign function; for the electrical parameter drift caused by temperature, define the medium - amplitude fluctuation loss , expressed as: ; ; where h is the step size constant; M is the upper limit of series truncation; n is the memory component series index; tanh(·) is the hyperbolic tangent function; for occasional short - circuit spikes of the sensor, a robust loss is defined , expressed as: ; .

[0014] Furthermore, the joint optimization parameter module makes the in the three - item loss independent of each other, and makes the fractional - order exponent of each loss term generated by an underlying variable , expressed as: ; where and are the fractional - order exponent and the underlying variable of the i - th loss respectively; and are the upper limit and the lower limit of the exponent respectively; during the update, the indirect - direct is adopted, and the update simultaneously updates and through backpropagation; the is used to map and update , and the overall loss of the healthy - condition current prediction model is defined, expressed as: ; the combined optimization is expressed as: ; where N is the total number of training samples; is the sample of the k - th frame; and are the parameters initialized finally.

[0015] Furthermore, the global adaptive activation function module uses in the three - item loss to be represented by , and respectively, and defines the global activation parameter , expressed as: ; and defines the global adaptive activation function of the healthy - condition current prediction model, expressed as: ; where is the global adaptive activation function; u is the input of the activation function.

[0016] Furthermore, the fault detection module, after training is completed, calculates the prediction residual for the newly input , and immediately evaluates the total fractional - order index , expressed as: , and uses the residual distribution under healthy conditions to determine the alarm threshold, expressed as: ; when , it is determined that a fault has occurred; among them, and are the mean and standard deviation under healthy working conditions respectively; is the safety factor.

[0017] The beneficial effects achieved by the present invention using the above solution are as follows:

[0018] (1) Aiming at the problems of the general electromagnetic fryer circuit fault detection system that the fault criterion is not flexible enough, once the working condition or environment changes, the missed detection or false detection rate increases; it is difficult to distinguish ordinary noise from tiny fault signals, resulting in poor fault detection accuracy. This solution constructs a healthy working condition current prediction model based on a fully connected feedforward neural network, designs loss functions for tiny resistance changes, parameter drift caused by temperature drift, and spike noise respectively, and generates parameters using underlying variable mapping to adaptively adjust the sensitivity of each loss; it can detect tiny coil turn-to-turn resistance changes within the noise level, is sensitive to medium-amplitude changes caused by temperature drift, and realizes the detection and suppression of sensor spike noise, ensuring both the timely detection of early and tiny faults and effectively avoiding false alarms.

[0019] (2) Aiming at the problems of the general electromagnetic fryer circuit fault detection system that it cannot capture the slight fluctuations caused by micro-changes in resistance and over-responds to sensor spike noise, resulting in poor fault detection effect. Based on designing loss functions for tiny resistance changes, parameter drift caused by temperature drift, and spike noise, this solution constructs a global adaptive activation function; the parameter self-adaptation is consistent with the sensitivity of the three types of losses, and the forward activation and backward loss are linked to amplify or suppress on the same fault feature scale, thereby improving the fault detection effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 is a schematic flow chart of an electromagnetic fryer circuit fault detection system based on artificial intelligence provided by the present invention.

[0021] The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0022] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0023] In the description of the present invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc. indicating the orientation or positional relationship are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation to the present invention.

[0024] Embodiment 1. Refer to Figure 1 , an electromagnetic deep fryer circuit fault detection system based on artificial intelligence provided by the present invention includes a data acquisition module, a pre - establishment module, a loss function design module, a joint optimization parameter module, a global adaptive activation function module, and a fault detection module;

[0025] The data acquisition module acquires the characteristic data of the electromagnetic deep fryer in a healthy state; and sends the data to the pre - establishment module;

[0026] The pre - establishment module constructs a feed - forward neural network model based on the standardized feature vectors in a healthy state; and sends the data to the loss function design module;

[0027] The loss function design module introduces three loss functions, respectively for tiny resistance changes, medium fluctuations, and spike noises; and sends the data to the joint optimization parameter module;

[0028] The joint optimization parameter module generates the parameters of the loss term through a learnable fractional - order exponential mapping and automatically adjusts them together with the network parameters during training through backpropagation; and sends the data to the global adaptive activation function module;

[0029] The global adaptive activation function module calculates the global activation parameters based on the fractional - order exponents of the three losses and constructs an adaptive activation function; and sends the data to the fault detection module;

[0030] The fault detection module performs fault detection on the newly input data.

[0031] Embodiment 2. Refer to Figure 1 , based on the above - mentioned embodiment, the data acquisition module selects key nodes at both ends of the power supply side and the inductance coil of the electromagnetic deep fryer to collect the voltage and current time - series signals in a healthy state; performs short - time Fourier transform on the time series to extract the frequency - domain and time - domain feature vectors; the time - domain statistical features include mean, variance, skewness, kurtosis, pulse factor, and waveform factor; the frequency - domain statistical features include spectral centroid, spectral bandwidth, spectral entropy, spectral flatness, and spectral flux; for the feature vectors in all dimensions perform zero - mean and unit - variance standardization to eliminate the differences in dimension and magnitude.

[0032] Embodiment 3, refer to Figure 1 , based on the above embodiment, the pre - establishment module establishes a model under healthy conditions for comparing with the actual signal to extract fault features; using the standardized feature vector as the input, a healthy operating condition current prediction model is constructed based on a fully - connected feed - forward neural network, expressed as: ; The residual is expressed as: ; where, is the predicted value of the k - th frame current; is the actual value; F(·) is the mapping of the fully - connected feed - forward neural network, are the network parameters; under normal operating conditions, the residual should be close to zero; faults such as coil short - circuit, open - circuit, and turn - to - turn short - circuit will produce significant outlier residuals.

[0033] Embodiment 4, refer to Figure 1 , based on the above embodiment, the loss function design module constructs a fractional - order loss function; using the short - memory characteristic of the fractional - order loss to make a more refined characterization of the residual, neither over - amplifying ordinary noise nor suppressing extreme transient spikes; dividing the residual - driven loss into three terms, corresponding to different fault modes respectively; for the resistance change of a tiny coil, a residual - sensitive detection loss is defined, expressed as: ; ; where, x is the input of the loss function, taking the residual ; c is a scaling constant; is the fractional - order exponent; is the gamma function; is the gradient; is the sign function; for the electrical parameter drift caused by temperature, a medium - amplitude fluctuation loss is defined, expressed as: ; ; where, h is the step constant; M is the upper limit of the series truncation; n is the series index of the memory component; tanh(·) is the hyperbolic tangent function; for the occasional short - circuit spikes of the sensor, a robust loss is defined, expressed as: ; .

[0034] Embodiment 5, refer to Figure 1 , based on the above embodiment, the joint optimization parameter module makes the in the three losses independent of each other. To avoid offline grid search, let the fractional - order exponent of each loss term be generated by a bottom - layer variable , expressed as: ; where, and They are the fractional - order exponent and the underlying variable of the \(i\) - th loss respectively; and They are the upper - bound exponent and the lower - bound exponent respectively; During the update, the indirect - direct is adopted. The update simultaneously updates and through back - propagation; updates through the mapping to ensure the correctness of the constrained optimization; Define the overall loss of the healthy - condition current prediction model , which is expressed as: ; where \(N\) is the total number of training samples; is the sample of the \(k\) - th frame; and are the parameters initialized with the final settings.

[0035] By performing the above operations, for the general electromagnetic fryer circuit fault - detection system, there are problems such as inflexible fault criteria. Once the working conditions or environment change, the missed - detection or false - detection rate increases; it is difficult to distinguish ordinary noise from tiny fault signals, resulting in poor fault - detection accuracy. This solution constructs a healthy - condition current prediction model based on a fully - connected feed - forward neural network, designs loss functions for tiny resistance changes, parameter drifts caused by temperature drifts, and spike noises respectively, and generates parameters through the mapping of underlying variables to adaptively adjust the sensitivity of each loss; it can detect tiny coil turn - to - turn resistance changes within the noise level, is sensitive to medium - amplitude changes caused by temperature drifts, and realizes the detection and suppression of sensor spike noises, ensuring both the timely detection of early and tiny faults and effectively avoiding false alarms.

[0036] Example 6, refer to Figure 1 , this example is based on the above - mentioned example. The global adaptive activation - function module is to enable the network to have adaptive expression ability among different fault - feature scales. The in the three losses are respectively represented by , and . Define the global activation parameter , which is expressed as: ; and define the global adaptive activation function of the healthy - condition current prediction model, which is expressed as: ; where is the global adaptive activation function; \(u\) is the input of the activation function; when becomes larger, also increases, the inflection point of the global adaptive activation function becomes steeper, and the network is better at learning high - frequency spike features; conversely, when becomes smaller, It also decreases, the activation is smoother, and the network is more inclined to fit large-scale steady changes, enabling the network to adaptively adjust its own non-linear expression ability for different feature scales when detecting tiny resistance changes, temperature drift, and spike short circuits.

[0037] By performing the above operations, aiming at the problem that the general electromagnetic fryer circuit fault detection system cannot capture the slight fluctuations caused by micro-resistance changes and over-responds to sensor spike noise, resulting in poor fault detection effect, this solution constructs a global adaptive activation function based on designing loss functions for tiny resistance changes, parameter drift caused by temperature drift, and spike noise; the parameter self-adaptation is consistent with the three types of loss sensitivities, and the forward activation and backward loss are linked to amplify or suppress on the same fault feature scale, thereby improving the fault detection effect.

[0038] Example Seven, refer to Figure 1 , based on the above example, after the training of the fault detection module is completed, for the newly input Calculate the prediction residual , and immediately evaluate the total fractional order index , expressed as: , determine the alarm threshold using the residual distribution under healthy conditions, expressed as: ; when , it is determined that a fault has occurred; where and are the mean and standard deviation under healthy conditions respectively; is the safety factor.

[0039] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device.

[0040] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention.

[0041] The above description of the present invention and its implementation manners is not restrictive. What is shown in the drawings is only one of the implementation manners of the present invention, and the actual structure is not limited thereto. In general, if those of ordinary skill in the art are inspired by it and, without departing from the gist of the present invention, design similar structural manners and embodiments to this technical solution without creative efforts, they shall fall within the protection scope of the present invention.

Claims

1. An electromagnetic fryer circuit fault detection system based on artificial intelligence, characterized in that: The system includes a data acquisition module, a pre - establishment module, a loss function design module, a joint optimization parameter module, a global adaptive activation function module, and a fault detection module; The data acquisition module acquires the electromagnetic fryer characteristic data under healthy conditions; The pre - establishment module constructs a feed - forward neural network model based on the standardized feature vectors under healthy conditions; The loss function design module introduces three loss functions, which are respectively for tiny resistance changes, medium fluctuations, and spike noises; The joint optimization parameter module generates the parameters of the loss term through a learnable fractional - order exponential mapping and automatically adjusts them together with the network parameters during training through backpropagation; The global adaptive activation function module calculates the global activation parameters based on the fractional - order exponents of the three losses and constructs an adaptive activation function; The fault detection module performs fault detection on the newly input data.

2. The electromagnetic deep fryer circuit fault detection system based on artificial intelligence according to claim 1, wherein: The loss function design module constructs a fractional-order loss function; by using the short memory property of the fractional-order loss to characterize the residual, the residual-driven loss is divided into three terms, corresponding to different fault modes respectively; for the resistance change of the micro coil, the residual-sensitive detection loss is defined , which is expressed as: ; ; where x is the input of the loss function, taking the residual ; c is the scaling constant; is the fractional-order exponent; is the gamma function; is the gradient; is the sign function; for the electrical parameter drift caused by temperature, the medium-amplitude fluctuation loss is defined , which is expressed as: ; ; where h is the step constant; M is the upper limit of the series truncation; n is the series index of the memory component; tanh(·) is the hyperbolic tangent function; for the occasional short-circuit spikes of the sensor, the robust loss is defined , which is expressed as: ; .

3. An electromagnetic deep fryer circuit fault detection system based on artificial intelligence according to claim 2, characterized in that: The combined optimization parameter module combines the three losses which are independent of each other, and sets the fractional order exponent of each loss term to be generated by an underlying variable , expressed as: ; where and are the fractional order exponent and the underlying variable of the i-th loss respectively; and are the upper and lower limits of the exponent respectively; during the update, indirect direct is adopted, and and are updated simultaneously through backpropagation; is used to map and update , and the overall loss of the healthy condition current prediction model is defined, expressed as: ; the combined optimization is expressed as: ; where N is the total number of training samples; is the sample of the k-th frame; and are the parameters finally set for initialization.

4. An electromagnetic deep fryer circuit fault detection system based on artificial intelligence according to claim 3, characterized in that: The global adaptive activation function module is to use the in the three losses respectively with , and to represent, define the global activation parameter , expressed as: ; and define the global adaptive activation function of the healthy condition current prediction model, expressed as: ; where is the global adaptive activation function; u is the activation function input.

5. The electromagnetic fryer circuit fault detection system based on artificial intelligence according to claim 4, characterized in that: The data acquisition module collects the voltage and current time series signals at the key nodes on the power supply side and both ends of the inductance coil from the electromagnetic fryer; performs short-time Fourier transform on the time series to extract the frequency domain and time domain feature vectors; performs zero-mean and unit-variance standardization on the feature vectors in all dimensions for standardization.

6. The electromagnetic fryer circuit fault detection system based on artificial intelligence according to claim 5 is characterized in that: The pre - established module is to establish a model under healthy conditions, using the standardized feature vectors as inputs, and constructing a healthy operating condition current prediction model based on a fully - connected feed - forward neural network, expressed as: ; The residual is expressed as: ; where is the predicted value of the current at the k - th frame; is the actual value; F(·) is the mapping of the fully - connected feed - forward neural network, are the network parameters.

7. An electromagnetic deep fryer circuit fault detection system based on artificial intelligence according to claim 6, characterized in that: The fault detection module, after training is completed, for newly input calculates the prediction residual , and instantaneously evaluates the total fractional order index , expressed as: , determines the alarm threshold using the residual distribution under healthy operating conditions, expressed as: ; when , it is determined that a fault has occurred; where and are the mean and standard deviation under healthy operating conditions respectively; is the safety factor.

Images