A small household appliance motor life prediction method based on data analysis

Abstract

The application relates to the technical field of data processing, and discloses a small household appliance motor life prediction method based on data analysis. The small household appliance motor only utilizes an original current sampling loop in a motor driving circuit, does not additionally add any vibration, temperature or special sensing device, and comprises the following steps: acquiring a working current time sequence signal output by the current sampling loop and synchronously acquiring a control signal of the motor driving circuit; pre-processing and feature extraction are performed on the working current time sequence signal to obtain a group of multi-dimensional features; the multi-dimensional features are input into a pre-trained health degradation evaluation model to obtain a health index of the small household appliance motor; and based on a historical sequence of the health index, a life prediction model is used to estimate the remaining service life of the small household appliance motor. The application realizes high-precision life prediction under zero hardware cost increment by deeply mining the value of the existing current signal.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and specifically to a method for predicting the lifespan of small household appliance motors based on data analysis. Background Technology

[0002] Currently, publicly available solutions for predicting motor lifespan generally rely on adding dedicated sensors to the motor. For example, vibration sensors are installed to collect vibration spectra to analyze mechanical wear, and temperature sensors are installed to monitor winding temperature rise to assess thermal aging. While these methods can achieve effective prediction, each additional sensor directly increases hardware material costs and complicates the installation process.

[0003] For low-cost, mass-produced small appliances such as vacuum cleaners and blenders, the selling price and profit margins are extremely limited. The cost of any additional sensors would be an unbearable burden on product competitiveness and commercial viability. Therefore, predictive solutions relying on additional sensors for vibration and temperature, while technically feasible, are impractical in the specific application of small appliances due to cost constraints.

[0004] In the current technology, there is no publicly available solution that can effectively predict the remaining lifespan of a motor in a small household appliance by utilizing only the existing and necessary current sampling circuit in the motor drive circuit, without adding any vibration, temperature, or dedicated sensing devices. This technological gap has prevented the small household appliance industry from integrating true predictive maintenance functions into its products for a long time.

[0005] Therefore, in order to solve the technical problem of predicting the lifespan of small household appliance motors without increasing hardware costs, this invention proposes a method for predicting the lifespan of small household appliance motors based on data analysis. Summary of the Invention

[0006] The purpose of this invention is to provide a method for predicting the lifespan of small household appliance motors based on data analysis, in order to solve the problem that the existing technology relies on the addition of vibration and temperature sensors, resulting in excessively high costs and making it unsuitable for low-cost small household appliances.

[0007] To solve the above-mentioned technical problems, the present invention specifically provides the following technical solution: A method for predicting the lifespan of small household appliance motors based on data analysis, wherein the small household appliance motor only utilizes the existing current sampling circuit in its motor drive circuit, without adding any vibration, temperature, or dedicated sensing devices, includes the following steps: Step S1: Obtain the working current timing signal output by the current sampling loop, and simultaneously obtain the control signal of the motor drive circuit; Step S2: Preprocess and extract features from the working current timing signal to obtain a set of multidimensional features, including a first subset of features for characterizing the mechanical state and a second subset of features for characterizing the thermal state; Step S3: Input the multidimensional features into the pre-trained health degradation assessment model to obtain the health indicators of small household appliance motors; Step S4: Based on the historical sequence of health indicators, estimate the remaining service life of the small household appliance motor using a life prediction model.

[0008] Compared with the prior art, the present invention has the following advantages: This invention completely abandons the traditional approach of relying on new vibration and temperature sensors. Instead, it creatively utilizes the existing current sampling loop in the motor drive circuit, which is used for control purposes, to obtain the core signal. This enables low-cost small household appliances to have predictive maintenance capabilities that were originally only found in high-end industrial equipment, without increasing material costs or making any structural changes. It fundamentally solves the core contradiction between "prediction accuracy and hardware cost," making it possible for this technology to be widely adopted in consumer products.

[0009] This invention, through specific signal processing and feature extraction steps, can deeply mine and separate rich feature subsets representing mechanical and thermal states from a single working current timing signal and its accompanying control signal. This is equivalent to "virtually" generating the functions of vibration and temperature sensors using algorithms, achieving a health state perception dimension comparable to multi-sensor solutions with zero hardware cost.

[0010] This invention introduces an attention-based fusion network and a cross-domain transfer learning training strategy, which can adaptively and dynamically fuse mechanical and thermal degradation information according to different operating conditions. It also overcomes the problem of scarce failure data for small household appliances, enabling the constructed health indicators to more accurately and robustly reflect the overall degradation degree of motor performance, thus laying a reliable foundation for subsequent life prediction.

[0011] This invention employs a hybrid prediction framework and designs intelligent decision-making rules that include a conservative early warning mode. When faced with prediction uncertainty, it can output reliable and conservative lifetime estimates and generate prompts that provide direct guidance to users. This not only improves the credibility and security of prediction results in practical applications but also greatly enhances the user experience and satisfaction of end users, transforming technical prediction into valuable user services.

[0012] The entire method and steps of this invention, from high-quality signal acquisition, advanced feature extraction, intelligent model evaluation to robust predictive decision-making, fully consider the limited computing and storage resources of the main controller of small home appliances. It can be deployed as a complete software algorithm package in the main controller of existing small home appliances without relying on cloud computing power, ensuring the real-time performance, low power consumption, data privacy and mass production feasibility of the final product. Attached Figure Description

[0013] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0014] Figure 1 This is a flowchart illustrating the method described in Embodiment 1 of the present invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] The concepts involved in this application will first be described with reference to the accompanying drawings. It should be noted that the following descriptions of various concepts are only for the purpose of making the content of this application easier to understand and do not constitute a limitation on the scope of protection of this application; furthermore, the embodiments and features in the embodiments of this application can be combined with each other unless otherwise specified. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0017] like Figure 1 As shown, this invention provides a method for predicting the lifespan of small household appliance motors based on data analysis. The method utilizes only the existing current sampling circuit in the motor drive circuit of the small household appliance motor, without adding any vibration, temperature, or dedicated sensing devices. The method includes the following steps: Step S1: Obtain the working current timing signal output by the current sampling loop, and simultaneously obtain the control signal of the motor drive circuit; Step S2: Preprocess and extract features from the working current timing signal to obtain a set of multidimensional features, including a first subset of features for characterizing the mechanical state and a second subset of features for characterizing the thermal state; Step S3: Input the multidimensional features into the pre-trained health degradation assessment model to obtain the health indicators of small household appliance motors; Step S4: Based on the historical sequence of health indicators, estimate the remaining service life of the small household appliance motor using a life prediction model.

[0018] Among them, small household appliance motors refer to drive motors used in various small household appliances, including but not limited to vacuum cleaner motors, blender motors, fan motors, water pump motors, etc. They are compact in structure, cost-sensitive, and usually do not have the space and conditions to install additional sensors.

[0019] A current sampling circuit refers to a current detection circuit originally configured in a motor drive circuit for overcurrent protection or current closed-loop control. It typically consists of a sampling resistor, an operational amplifier, an analog-to-digital converter, etc., and its output is a voltage signal or digital signal that is proportional to the motor's operating current.

[0020] The operating current timing signal refers to the sequence of current data continuously collected by the current sampling circuit and arranged in chronological order during motor operation. It contains rich information about the electromagnetic, mechanical, and thermal characteristics of the motor.

[0021] Control signals refer to electrical signals used in motor drive circuits to control the operating status of the motor, including but not limited to pulse width modulation duty cycle signals, drive enable signals, and direction signals, which can reflect the real-time operating conditions of the motor.

[0022] Feature extraction refers to the process of extracting quantitative indicators that reflect the health status of the motor from the preprocessed operating current time-series signal. The result is usually represented as a set of multidimensional feature vectors.

[0023] The first type of feature subset refers to the feature set used to characterize the mechanical state of a motor. It mainly comes from the ripple component related to mechanical rotation in the working current timing signal and can reflect mechanical degradation phenomena such as bearing wear, rotor imbalance, and mechanical loosening.

[0024] The second type of feature subset refers to the feature set used to characterize the thermal state of the motor. It is mainly based on the winding temperature rise and thermal characteristic parameters estimated by the operating current and control signal, and can reflect thermal degradation phenomena such as winding insulation aging and thermal fatigue.

[0025] A health degradation assessment model is a model based on machine learning or deep learning. Its input is multidimensional features, and its output is a quantitative indicator that can characterize the overall health status of the motor. This model is usually trained with a large amount of historical data.

[0026] Health indicators are single numerical values ​​or vectors output by health degradation assessment models to quantitatively describe the current health status of a motor. Their values ​​have a clear monotonic relationship with the remaining lifespan of the motor and can be used for subsequent lifespan prediction.

[0027] A lifespan prediction model is a model used to predict the remaining lifespan of a motor based on the historical trends of health indicators. It can be an analytical model based on the degradation process or a machine learning model based on time series data.

[0028] Remaining service life refers to the length of time a motor is expected to continue operating normally under its current health condition. It can be measured in hours or in the number of operating cycles.

[0029] In one specific implementation, the small appliance motor is a high-speed brushless DC motor for vacuum cleaners, and its drive circuit is equipped with a sampling resistor for overcurrent protection and an analog-to-digital conversion channel to form a current sampling loop.

[0030] During motor operation, the current sampling circuit continuously collects the operating current at a fixed sampling frequency, forming an operating current timing signal. Simultaneously, the pulse width modulation duty cycle signal output by the controller is synchronously collected and used as a control signal.

[0031] The acquired operating current timing signal is preprocessed by filtering and normalization to remove high-frequency noise and abnormal spikes. Subsequently, the preprocessed current signal is synchronously demodulated to separate the current ripple component that is synchronized with the mechanical rotation of the rotor. The harmonic amplitude and amplitude ratio related to the mechanical state in this component are extracted by order tracking analysis to form the first type of feature subset.

[0032] Based on the parameters on the motor nameplate and the thermal resistance, thermal capacity, etc. obtained from experimental calibration, a first-order equivalent thermal network model of the winding is established. The Joule thermal power is calculated based on the effective value of the working current, and the switching loss power is calculated based on the pulse width modulation duty cycle. The two are superimposed as the input of the thermal network model, and the winding temperature rise curve is obtained by solving. The thermal time constant, model fitting error, and peak-to-peak temperature rise features are extracted from the temperature rise curve to form a second type of feature subset.

[0033] The first and second feature subsets are input into a pre-trained health degradation assessment model, which consists of two encoding networks and an attention fusion module. The network encodes mechanical and thermal features respectively, and then dynamically adjusts the fusion weights based on the current effective value and duty cycle to output a health index between 0 and 1.

[0034] As motor operating time increases, this health indicator gradually declines. The historical sequence of the health indicator's changes over time is input into the lifespan prediction model. This model simultaneously employs an exponential degradation model and a lightweight temporal neural network for prediction, obtaining two remaining lifespan estimates. These estimates are then fused based on goodness of fit and prediction uncertainty to output the final remaining lifespan. When the remaining lifespan falls below a preset threshold, the system issues a maintenance warning, prompting the user to replace the motor or perform maintenance in a timely manner.

[0035] As one embodiment of the present invention, step S1 includes: S11: Acquire the operating current timing signal in a way that is synchronized with the edge trigger of the pulse width modulation signal, wherein sampling is performed at the midpoint of the turn-on or turn-off period in each carrier cycle, and the sampling frequency is not less than 20 times the current electrical frequency. S12: Synchronously acquire the control signal of the motor drive circuit, and reconstruct the electrical angle and load condition information of the motor in real time based on the control signal. The control signal includes a pulse width modulation duty cycle signal, and the electrical angle is used to demodulate the working current timing signal in sync with the mechanical rotation.

[0036] Among them, edge-triggered synchronization of pulse width modulation signal means that the start time of current sampling is kept in a fixed time relationship with the rising or falling edge of pulse width modulation signal, so that each sampling falls at the same relative position within the period of pulse width modulation signal, thereby ensuring that the sampling results have a consistent phase reference for different duty cycles and different switching states.

[0037] The carrier period refers to a complete cycle of the carrier waveform in a pulse width modulation signal, which determines the switching frequency of the switching device.

[0038] The on-time period refers to the time interval during which a switching device is in the on state within one carrier cycle.

[0039] The off period refers to the time interval during which a switching device is in the off state within one carrier cycle.

[0040] Midpoint sampling refers to sampling at the center of the time axis during the turn-on or turn-off period to avoid current spikes and transient processes at the moment of switching action, thereby improving sampling stability and signal-to-noise ratio.

[0041] Electrical frequency refers to the fundamental frequency of the current in the stator winding of a motor. It is related to the motor speed and the number of pole pairs, and reflects the electrical operating state of the motor.

[0042] Electrical angle refers to the angular quantity that describes the change of the spatial position of the stator magnetic field of a motor over time. It changes synchronously with the motor speed and is used to perform signal processing of current signals that are synchronized with mechanical rotation.

[0043] Load condition information refers to the operating status information related to the current output torque, speed, and power of the motor, which can be indirectly reflected by the changing trend of the control signal and the magnitude of the duty cycle.

[0044] Demodulation synchronized with mechanical rotation refers to using electrical angle as a reference to decompose the working current timing signal into components related to the rotor's mechanical rotation cycle, thereby extracting information related to changes in the mechanical state.

[0045] In one specific implementation, the motor drive circuit employs a three-phase inverter structure based on pulse width modulation (PWM). The PWM signal is generated by the controller and output to the power switching devices. The current sampling loop consists of a sampling resistor and an operational amplifier, and its output is converted into a digital signal by an analog-to-digital converter (ADC). The controller triggers a sampling timer on the rising edge of the PWM signal. The timer generates a sampling trigger signal at the midpoint of the on-time of each carrier cycle, driving the ADC to sample the current. The sampling frequency is dynamically adjusted according to the current electrical frequency, always maintaining a value at least 20 times the electrical frequency.

[0046] While sampling the current, the controller synchronously reads the current pulse width modulation duty cycle signal and calculates the real-time electrical angle of the motor based on the duty cycle change. The electrical angle is continuously updated with changes in duty cycle and motor speed, and is used for subsequent synchronous demodulation of the operating current timing signal to extract the current ripple component related to rotor mechanical rotation. Load condition information is indirectly obtained through the magnitude and rate of change of the duty cycle; an increased duty cycle indicates an increased load, and a decreased duty cycle indicates a decreased load. In this way, the operating current timing signal and the control signal are strictly synchronized in time, providing accurate phase and operating condition references for subsequent feature extraction.

[0047] In one embodiment of the present invention, in step S2, the first subset of features used to characterize the mechanical state is obtained through the following steps: The working current timing signal is synchronously demodulated to separate the current ripple component that is synchronized with the mechanical rotation frequency of the small household appliance motor rotor. The current ripple component contains sideband components centered on the electrical frequency and its harmonics, with the offset being the mechanical rotation frequency. Order tracking analysis was performed on the current ripple component to calculate the amplitude of its fundamental frequency integer multiple harmonics. The order tracking analysis adopted an order tracking algorithm based on the Volk-Kalman filter. Select the amplitudes of two target harmonics from the results of the order tracking analysis, and calculate the amplitude ratio between different target harmonics; The selected amplitude and the calculated ratio are combined to form the first type of feature subset.

[0048] Furthermore, the synchronous demodulation process is implemented using orthogonal demodulation technology, including the following steps: Based on the pulse width modulation duty cycle signal, the real-time electrical angle signal of the motor is reconstructed as a demodulation reference. Using the electrical angle signal as the phase reference, the working current timing signal is decomposed into synchronous and quadrature components; The synchronous component and the quadrature component are combined into a complex current ripple signal. The instantaneous amplitude and instantaneous phase of the current ripple signal are calculated from the synchronous component and the quadrature component, respectively.

[0049] Among them, synchronous demodulation processing refers to using a phase reference synchronized with the motor rotation to separate the modulation components related to mechanical rotation from the working current timing signal, so as to reveal the mechanical information that was originally masked by high-frequency switching ripple and electromagnetic noise.

[0050] The current ripple component refers to the periodic fluctuation part of the operating current other than the DC or fundamental frequency component. It includes frequency components generated by the combined effects of the electromagnetic and mechanical characteristics of the motor.

[0051] Mechanical rotational frequency refers to the number of revolutions per second of the motor rotor. It directly corresponds to the actual speed of the motor and is an important frequency reference reflecting the mechanical state.

[0052] Sideband components refer to frequency components that appear on both sides of the electrical frequency and its harmonics. Their offset is equal to the mechanical rotation frequency or a multiple thereof, and is a typical manifestation of mechanical modulation on current signals.

[0053] Order tracking analysis refers to resampling the signal according to the rotation angle under the condition of rotational speed change, so that the component synchronized with the rotational speed is converted into a fixed frequency, thereby accurately extracting the harmonic features related to the rotational order.

[0054] The fundamental frequency integer multiples of harmonics refer to components whose frequency is an integer multiple of the mechanical rotation frequency. Their amplitude changes can reflect mechanical degradation phenomena such as rotor imbalance and bearing wear.

[0055] The Vold-Kalman filter is a tracking filter suitable for non-stationary signal analysis. It can accurately extract frequency components of a specific order under varying rotational speed conditions and has good noise immunity and tracking capability.

[0056] The target harmonic refers to a specific order component that is sensitive to mechanical degradation and is selected from the order tracking results. It usually includes the fundamental frequency and several harmonic components.

[0057] The amplitude ratio refers to the relative proportion between the amplitudes of different target harmonics. Its change can reflect the changes in the characteristics of the mechanical system and is more robust than a single amplitude.

[0058] Quadrature demodulation (QDM) technology refers to decomposing a signal into two mutually orthogonal components, and extracting the instantaneous amplitude and instantaneous phase of the signal through phase alignment and synthesis. It is a commonly used implementation method of synchronous demodulation.

[0059] The electrical angle signal is an angular quantity that describes the change of the spatial position of the stator magnetic field of a motor over time. It is synchronized with the motor speed and is a key reference signal for achieving synchronous demodulation.

[0060] Synchronization component refers to the component in the operating current timing signal that is in phase with the electrical angle signal, reflecting the electromagnetic and mechanical modulation information that is synchronized with the rotation of the motor.

[0061] The quadrature component refers to the component in the working current timing signal that is 90 degrees out of phase with the electrical angle signal. Together with the synchronization component, it constitutes a complex current ripple signal.

[0062] The complex form of the current ripple signal refers to a complex signal constructed by taking the synchronous component and the quadrature component as the real part and the imaginary part, respectively. Its magnitude is the instantaneous amplitude and its phase angle is the instantaneous phase, which can completely describe the modulation characteristics of the current ripple.

[0063] Instantaneous amplitude refers to the magnitude of the current ripple signal at a certain moment, and its change can reflect the dynamic characteristics and degradation degree of the mechanical system.

[0064] Instantaneous phase refers to the phase state of the current ripple signal at a certain moment. Its changes can reflect the uniformity of mechanical rotation and load fluctuations.

[0065] In one specific implementation, the operating current timing signal is first low-pass filtered to remove high-frequency switching noise, retaining the mid-to-low frequency components related to mechanical rotation. Based on the pulse width modulation duty cycle signal and the estimated motor speed, a continuously varying electrical angle signal is reconstructed. Using the electrical angle signal as a reference, the operating current timing signal is multiplied by sine and cosine reference signals respectively to obtain synchronization and quadrature components. After low-pass filtering, the synchronization and quadrature components are synthesized into a complex current ripple signal, whose instantaneous amplitude and phase change with the motor's mechanical state. Order tracking analysis based on a Vold-Kalman filter is performed on the current ripple signal to extract the mechanical rotation frequency and the amplitude of its harmonic components. The fundamental frequency and second harmonic are selected as target harmonics, and their amplitudes and amplitude ratios are calculated. As bearing wear or rotor imbalance worsens, the ratio of the second harmonic amplitude to the fundamental frequency amplitude gradually increases; therefore, this ratio can effectively characterize the degree of mechanical degradation. The fundamental frequency amplitude, the second harmonic amplitude, and the ratio between the two are combined to form the first type of feature subset, which is used as input for the subsequent health degradation assessment model.

[0066] In one embodiment of the present invention, in step S2, the second subset of features used to characterize the thermal state is obtained through the following steps: Based on the thermal parameters obtained through the calibration of small household appliance motor models, a first-order equivalent thermal network model of the winding is established. Calculate the thermal power of the input model. The thermal power is the superposition of Joule thermal power and switching loss power. Joule thermal power is determined by the effective value of the operating current and the winding resistance, while switching loss power is determined by the pulse width modulation duty cycle signal and the preset switching loss coefficient. The thermal power is input into the first-order equivalent thermal network model, and the estimated temperature rise curve of the winding is obtained in real time. Features for characterizing fatigue accumulation of insulating materials under thermal cycling stress are extracted from the temperature rise curves, and the extracted features are used as a second subset of features, including thermal inertia features, model mismatch features, and thermal fluctuation features.

[0067] Furthermore, the thermal inertia characteristic is the time it takes for the estimated temperature rise of the winding to reach 63.2% of its steady-state value from the initial value, which is the time constant of the first-order equivalent thermal network model; The model mismatch characteristic is the coefficient of determination or root mean square error of the fit between the temperature rise curve data points obtained by real-time solution and the theoretical curve generated based on the time constant and the first-order step response model. Thermal fluctuation characteristics are the difference between the peak and trough values ​​of the estimated temperature rise of the windings during a complete and typical working cycle of a small household appliance motor, i.e., the peak-to-peak temperature rise.

[0068] Thermal parameters refer to physical quantities used to describe the heat conduction and heat storage capacity of motor windings, including thermal resistance and thermal capacity. Their values ​​are determined through experimental calibration or motor parameters provided by the manufacturer.

[0069] The first-order equivalent thermal network model simplifies the motor windings and their heat dissipation path into a thermal network consisting of thermal resistance and thermal capacity connected in series, and is used to describe the dynamic process of winding temperature changing with thermal power.

[0070] Thermal power refers to the amount of heat input into the winding per unit time, which is the fundamental reason for the increase in winding temperature.

[0071] Joule heat power refers to the heat loss generated when current passes through the winding resistance, and its magnitude is proportional to the square of the current and the resistance.

[0072] The effective value of the operating current is the square root of the average value of the square of the current over one cycle, and it is a key parameter for calculating Joule thermal power.

[0073] The winding resistance refers to the DC resistance of the motor stator winding. It changes slightly with temperature, but in this method, it can be approximated based on the calibrated value.

[0074] Switching power loss refers to the heat loss generated by power switching devices during the switching process, and its magnitude is related to the switching frequency, duty cycle and device characteristics.

[0075] The switching loss factor is a proportionality coefficient used to quantify the relationship between switching loss and duty cycle; its value is obtained through experimental calibration.

[0076] The winding temperature rise curve is a curve of winding temperature change over time calculated by a first-order equivalent thermal network model, reflecting the dynamic thermal response of the winding under load changes.

[0077] Thermal cycling stress refers to the thermal fatigue effect experienced by windings during repeated heating and cooling, and is an important factor leading to the aging of insulation materials.

[0078] Fatigue accumulation in insulating materials refers to the process by which the performance of insulating materials gradually declines under long-term thermal cycling. The degree of accumulation is related to the temperature rise, the frequency of change, and the duration.

[0079] Thermal inertia refers to the degree to which the temperature change of the winding lags behind the change in thermal power. It is reflected by the time constant of the thermal network. The larger the time constant, the stronger the thermal inertia.

[0080] The time constant refers to the time required for a first-order system to reach 63.2% of its final value under a step input, and it is an important parameter describing the dynamic response speed of the system.

[0081] Model mismatch characteristics refer to the degree of deviation between the actual temperature rise curve and the theoretical model. An increase in this deviation indicates a change in the thermal characteristics of the winding, which may be caused by insulation aging or changes in heat dissipation conditions.

[0082] The coefficient of determination is an indicator used to measure how well a model fits the data; the closer its value is to 1, the better the fit.

[0083] The root mean square error (RMSE) is an index used to measure the average deviation between the model's predicted values ​​and the actual values. The smaller the value, the better the fit.

[0084] Thermal fluctuation characteristics refer to the range of temperature change of the winding within one working cycle, reflecting the severity of thermal cycling, and are an important indicator for assessing insulation thermal fatigue.

[0085] Peak-to-peak temperature rise refers to the difference between the maximum and minimum winding temperature within one working cycle, and is the most direct parameter for describing temperature fluctuations.

[0086] In one specific implementation, the winding temperature change under different operating currents is first measured experimentally. The thermal resistance and thermal capacity of the small household appliance motor are obtained using curve fitting, establishing a first-order equivalent thermal network model. During motor operation, the effective value of the current is calculated based on the operating current timing signal, and the Joule thermal power is obtained by combining it with the winding resistance. Simultaneously, the switching loss power is calculated based on the pulse width modulation duty cycle signal and a pre-calibrated switching loss coefficient, and the two are added together to obtain the real-time thermal power. The thermal power is input into the first-order equivalent thermal network model, and the estimated winding temperature rise curve is obtained through numerical integration. Thermal inertia characteristics, i.e., the time required for the temperature rise to reach 63.2% of the steady-state value, are extracted from the temperature rise curve. By comparing the estimated temperature rise curve with the theoretical curve of the first-order step response based on the time constant, the coefficient of determination or root mean square error is calculated as a model mismatch characteristic. Within a complete operating cycle, the maximum and minimum temperatures are recorded, and the difference between them is calculated as a thermal fluctuation characteristic. The thermal inertia characteristic, model mismatch characteristic, and thermal fluctuation characteristic are combined to form a second subset of features, which is used as input for the subsequent health degradation assessment model. As the winding insulation ages, the increased thermal resistance leads to a longer time constant and a greater model mismatch. At the same time, the thermal fluctuation characteristics may also increase due to the decrease in heat dissipation capacity. Therefore, the second type of feature subset can effectively characterize the thermal state degradation of the motor.

[0087] As one embodiment of the present invention, step S3 includes: S31: Input the first type of feature subset into the first coding network for processing to obtain the mechanical degradation latent variable; S32: Input the second type of feature subset into the second coding network for processing to obtain thermal degradation latent variables; S33: Obtain a query signal reflecting the current motor operating condition. The query signal includes the effective value of the current operating current, the pulse width modulation duty cycle, and the real-time speed reconstructed based on the control signal. S34: Based on the query signal, the fusion weights of the mechanical degradation latent variables and the thermal degradation latent variables are dynamically calculated through the attention fusion module, and weighted fusion is performed to obtain a comprehensive representation; S35: Map the comprehensive representation to a one-dimensional health indicator.

[0088] Furthermore, the first and second encoding networks are obtained through cross-domain transfer learning, including: A basic feature extraction network is pre-trained on a public multi-sensor dataset containing various types of industrial motors, including vibration, temperature, and current signals. The parameters of the basic feature extraction network are fixed, and its original input and output layers related to vibration and temperature sensors are removed. An adaptation layer is constructed to map the output dimension of the basic feature extraction network to the single current feature space of the small household appliance motor; Using a dataset of small household appliance motors containing single current signals, the adaptation layer and subsequent regression layer are jointly fine-tuned to obtain a health degradation assessment model.

[0089] Among them, the first encoding network refers to the neural network used to compress and abstract the features of the first type of feature subset. Its core function is to extract core information that can characterize the degree of mechanical degradation from the mechanical state features.

[0090] Mechanical degradation latent variables refer to the output of the first coding network. They are low-dimensional abstract representations of mechanical state characteristics and can concisely and comprehensively reflect the degradation state of the motor mechanical system.

[0091] The second coding network refers to a neural network used to compress and abstract features from the second type of feature subset. Its core function is to extract core information that can characterize the degree of thermal degradation from the thermal state features.

[0092] Thermal degradation latent variables refer to the output of the second coding network. They are low-dimensional abstract representations of thermal state characteristics and can concisely and comprehensively reflect the degradation state of the motor thermal system.

[0093] Real-time speed refers to the current rotational speed of the motor obtained by reconstructing the control signal of the motor drive circuit, which can directly reflect the motor's load and operating status.

[0094] The attention fusion module refers to a fusion unit that can dynamically adjust the weights of different latent variables according to the query signal. Its core function is to enable the fusion process to adapt to different operating conditions of the motor and improve the accuracy of the comprehensive representation.

[0095] Weighted fusion refers to multiplying two latent variables by their respective fusion weights and then summing them to obtain a unified representation that comprehensively reflects the mechanical and thermal degradation states.

[0096] Comprehensive characterization refers to the weighted and fused low-dimensional feature vector, which can take into account the degradation information of both the mechanical and thermal dimensions of the motor, providing a comprehensive basis for the generation of health indicators.

[0097] Cross-domain transfer learning refers to a learning method that transfers knowledge and model parameters learned from one domain (multi-sensor scenario of industrial motors) to another related domain (single-current scenario of small household appliance motors). Its core function is to solve the problem of insufficient labeled data for small household appliance motors.

[0098] Multi-sensor public datasets refer to publicly available datasets that contain various types of industrial motors and simultaneously collect signals from multiple sources such as vibration, temperature, and current. These datasets are used to provide rich training samples for basic feature extraction networks.

[0099] The basic feature extraction network refers to a neural network pre-trained on a multi-sensor public dataset that can extract multi-dimensional degradation features of a motor and has strong feature extraction capabilities.

[0100] Fixing the parameters of the basic feature extraction network means keeping the parameters of each layer of the pre-trained basic network unchanged during the transfer learning process, and only adjusting the parameters of the newly added network layers to retain the general feature extraction capabilities learned by the basic network.

[0101] The adaptation layer is a transitional network layer used to connect the basic feature extraction network and the subsequent regression layer. Its core function is to convert the feature dimensions output by the basic network into dimensions that fit the single current feature space of the small household appliance motor, thereby achieving domain adaptation.

[0102] The single current feature space refers to the feature space composed solely of features extracted from current signals, and is a dedicated space for feature representation in the context of small household appliance motors.

[0103] The small household appliance motor dataset refers to a dataset specifically collected for small household appliance motors, containing only single current signals and corresponding health status labels, used for fine-tuning the adaptation and regression layers.

[0104] Joint fine-tuning refers to adjusting the parameters of the adaptation layer and subsequent regression layer simultaneously during training, so that the entire network can better adapt to the single-current scenario of small household appliance motors and improve the prediction accuracy of health indicators.

[0105] The regression layer refers to the network layer in the health degradation assessment model used to map the comprehensive representation into a one-dimensional health indicator. Its core function is to realize the transformation from low-dimensional features to quantitative values ​​of health status.

[0106] In one specific implementation, both the first and second encoding networks employ a three-layer fully connected neural network structure. The first encoding network takes a first-class feature subset as input, and after linear transformation and activation function processing, outputs a mechanical degradation latent variable with a dimension of 16. The second encoding network takes a second-class feature subset as input and similarly outputs a thermal degradation latent variable with a dimension of 16. The query signals are obtained through real-time calculation. The effective value of the operating current is calculated from the current timing signal, the pulse width modulation duty cycle is synchronously obtained from the motor drive circuit, and the real-time speed is reconstructed based on the duty cycle signal and motor parameters. The attention fusion module receives three query signal parameters and calculates two fusion weights through a fully connected network. The sum of these weights is 1. Under high load conditions, the mechanical degradation latent variable has a higher weight, while under long-term operation conditions, the thermal degradation latent variable has a higher weight. The two latent variables are multiplied by their corresponding weights and summed to obtain a comprehensive representation with a dimension of 16. This comprehensive representation is input into the regression layer, and after linear transformation, outputs a one-dimensional health index between 0 and 1. The closer the value is to 1, the better the motor's health; the closer it is to 0, the more severe the degradation.

[0107] In the cross-domain transfer learning process, a publicly available multi-sensor dataset of industrial motors was first selected. This dataset contains vibration, temperature, and current signals of more than ten types of industrial motors, along with corresponding health status labels. A basic feature extraction network was pre-trained based on this dataset. This network employs a convolutional neural network structure, which can effectively extract degradation features from multi-source signals. After pre-training, the parameters of all convolutional and pooling layers of the network were fixed, and the input layers for receiving vibration and temperature signals, as well as their corresponding output branches, were removed, retaining only the input correlation structure for receiving current signals and the intermediate feature output layer. A fully connected adaptation layer was constructed to map the 64-dimensional features output by the basic network to 32-dimensional features, adapting to the single-current feature space of small household appliance motors. Subsequently, a single-current dataset of small household appliance motors was collected. This dataset contains the full-lifecycle current signals and corresponding health labels of 50 small household appliance motors from normal to failure states. This dataset was used to jointly fine-tune the adaptation layer and subsequent regression layers. During fine-tuning, a mean squared error loss function was used to minimize the error between the model's output health indicators and the actual health status. After fine-tuning, the final health degradation assessment model was obtained. This model can make full use of the rich information in industrial motor datasets, while also adapting to single-current scenarios for small household appliance motors, effectively improving the prediction accuracy of health indicators.

[0108] As one embodiment of the present invention, step S4 includes: S41: Based on the start-stop status of small household appliance motors, filter health indicators under stable operating conditions and construct historical sequences for prediction; S42: Input the historical sequence into the exponential degradation model and the lightweight time-series neural network simultaneously to obtain the first lifetime estimate and its goodness of fit output by the exponential degradation model, and the second lifetime estimate and its prediction uncertainty output by the time-series neural network. S43: Based on the goodness of fit and prediction uncertainty, select the first lifetime estimate and the second lifetime estimate, and output the final remaining lifetime estimate.

[0109] Furthermore, the step of selecting between the first lifetime estimate and the second lifetime estimate includes: Compare the goodness of fit with a preset confidence threshold, and the prediction uncertainty with a preset uncertainty threshold; If the goodness of fit is higher than its threshold and the prediction uncertainty is lower than its threshold, it indicates that the two models predict in a consistent and reliable manner, and the weighted average of the two models is output as the final remaining lifetime. If the goodness of fit is lower than its threshold or the prediction uncertainty is higher than its threshold, it indicates that the current prediction uncertainty is high. The conservative warning mode is activated, and the shorter lifespan value predicted in the dual model is output as the remaining lifespan estimate. An auxiliary prompt message "low prediction reliability, it is recommended to shorten the inspection cycle" is generated.

[0110] Among them, the start-stop state refers to the working state of the small household appliance motor when it is starting, running or stopping. It can be determined by the control signal of the motor drive circuit. The health indicators during the start-up and stop phases are unstable and need to be eliminated.

[0111] Stable operation refers to a state in which parameters such as motor speed, current, and load remain stable without drastic fluctuations. Health indicators under this state can accurately reflect the degree of motor degradation.

[0112] Historical sequence refers to a data sequence formed by arranging health indicators under stable operating conditions in chronological order after screening. It contains trend information on the change of motor health status over time and is the core input for life prediction.

[0113] The exponential degradation model is an analytical model built on the assumption that motor health indicators degrade exponentially over time. Its core function is to estimate the remaining service life by fitting the degradation trend through historical sequences.

[0114] Lightweight temporal neural networks are time-series prediction networks with simplified structures and low computational cost, suitable for small household appliance scenarios. They can capture the nonlinear variation characteristics in the historical sequence of health indicators and achieve high-precision lifespan estimation.

[0115] The first life estimate refers to the prediction of the remaining lifespan of the motor by the exponential degradation model, which is calculated based on the exponential degradation trend of health indicators.

[0116] Goodness of fit refers to the degree of agreement between the predicted values ​​of an exponential degradation model and the actual values ​​of historical health indicators; its magnitude reflects the quality of the model's fit.

[0117] The second life estimate refers to the prediction of the remaining lifespan of a motor by a lightweight temporal neural network, which is derived from learning the nonlinear variation characteristics of historical health indicators.

[0118] Prediction uncertainty refers to the reliability of the lifetime estimate output by a lightweight temporal neural network. Its magnitude reflects the fluctuation range and credibility of the prediction result.

[0119] The remaining useful life estimate refers to the final output value, which characterizes the length of time the motor can still work normally under its current health condition. It is the ultimate goal of life prediction.

[0120] The confidence threshold is a preset critical value used to judge whether the goodness of fit is acceptable. It is calibrated based on a large amount of experimental data and is used to measure the fitting effect of the exponential degradation model.

[0121] Uncertainty threshold refers to a preset critical value used to determine whether the uncertainty of prediction is acceptable. It is calibrated according to the application scenario requirements of small household appliance motors and is used to measure the reliability of the prediction results of time-series neural networks.

[0122] Consistent predictions between the two models mean that the lifetime estimates output by the exponential degradation model and the lightweight temporal neural network are similar, and their respective reliability indices meet the requirements, indicating that the prediction results are consistent.

[0123] The weighted average is the average value calculated by assigning different weights to two lifetime estimates based on their reliability. The weighting is positively correlated with the goodness of fit and prediction uncertainty, resulting in a higher proportion of estimates with higher reliability.

[0124] High prediction uncertainty means that the prediction uncertainty of the time series neural network exceeds the preset threshold, or the goodness of fit of the exponential degradation model is lower than the preset threshold, indicating that the reliability of the current prediction results is insufficient.

[0125] The conservative early warning mode refers to a more cautious strategy to output a lifespan estimate when the reliability of the prediction is insufficient, prioritizing user safety and avoiding risks caused by motor failure due to prediction deviation.

[0126] The shorter lifespan value refers to the smaller of the two lifespan estimates output by the exponential degradation model and the lightweight temporal neural network. Choosing this value can minimize the risk of premature motor failure.

[0127] The auxiliary prompt information refers to the text prompts generated when the conservative early warning mode is activated. They are used to inform users of the reliability of the current prediction results and provide targeted maintenance suggestions to help users take timely measures.

[0128] In one specific implementation, the start-stop status of the small household appliance motor is first determined by the pulse width modulation duty cycle signal and the working current timing signal of the motor drive circuit. When the duty cycle is stable within the preset range and the current fluctuation amplitude is less than the set value, the motor is determined to be in a stable operating state. Health indicators in this state are filtered and arranged in chronological order to construct a historical sequence. The length of the historical sequence is set to the health indicator data of the most recent 100 stable operating moments.

[0129] The constructed historical sequence is simultaneously input into the exponential degradation model and the lightweight temporal neural network. The exponential degradation model outputs a first lifespan estimate by fitting the historical sequence to an exponential curve and calculates the goodness of fit, which is represented by the coefficient of determination. The closer the value is to 1, the better the fit. The lightweight temporal neural network adopts a two-layer gated recurrent unit structure. After inputting the historical sequence, it learns the time evolution of health indicators, outputs a second lifespan estimate, and calculates the prediction uncertainty. The prediction uncertainty is represented by the variance of the prediction results. The smaller the variance, the more reliable the prediction.

[0130] The confidence threshold is preset to 0.8, and the uncertainty threshold to 5 hours. The goodness of fit of the exponential degradation model is compared with the confidence threshold, and the prediction uncertainty of the lightweight temporal neural network is compared with the uncertainty threshold. If the goodness of fit is 0.85, which is higher than the confidence threshold of 0.8, and the prediction uncertainty is 3 hours, which is lower than the uncertainty threshold of 5 hours, it indicates that the two models predict consistently and reliably. In this case, weights are assigned according to the goodness of fit and the prediction uncertainty, with higher weights for higher goodness of fit and lower prediction uncertainty. The weighted average of the first lifetime estimate and the second lifetime estimate is calculated, and this average is used as the final remaining lifetime estimate.

[0131] If the goodness of fit is 0.7, which is lower than the confidence threshold of 0.8, or the prediction uncertainty is 6 hours, which is higher than the uncertainty threshold of 5 hours, it indicates that the current prediction uncertainty is high. In this case, a conservative warning mode is activated. The first life estimate and the second life estimate are compared, and the smaller value is selected as the final remaining life estimate. At the same time, an auxiliary prompt message "Low prediction reliability, it is recommended to shorten the inspection cycle" is generated. This prompt message can be pushed to the user through the display panel of the small appliance or the associated mobile APP to remind the user to check and maintain the motor in a timely manner to avoid premature motor failure.

[0132] Throughout the process, the exponential degradation model is responsible for capturing the linear degradation trend of health indicators, while the lightweight temporal neural network is responsible for capturing nonlinear degradation features. The combination of the two models and the dynamic selection strategy not only ensures the prediction accuracy but also improves the reliability and security of the prediction results, making it suitable for the actual application scenarios of small household appliance motors.

[0133] It should be noted that the terminology used in this application is for the purpose of describing specific embodiments only and is not intended to limit the scope of this application. As shown in the specification and claims of this application, unless the context clearly indicates otherwise, words such as "a," "an," "an," and / or "the" do not specifically refer to the singular and may also include the plural. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, or apparatus that includes said element.

[0134] It should also be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation on this application. Unless otherwise expressly specified and limited, the terms "installed," "connected," "linked," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication between two elements. For those skilled in the art, the specific meaning of the above terms in this application can be understood according to the specific circumstances.

[0135] The embodiments and / or implementation methods described above are merely preferred embodiments and / or implementation methods for implementing the technology of the present invention, and are not intended to limit the implementation methods of the technology of the present invention in any way. Any person skilled in the art can make some modifications or alterations to other equivalent embodiments without departing from the scope of the technical means disclosed in the content of the present invention, but they should still be regarded as the technology or embodiments that are substantially the same as the present invention.

[0136] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. The above descriptions are only preferred embodiments of this application. It should be noted that due to the limitations of written expression, while there are objectively infinite specific structures, those skilled in the art can make several improvements, modifications, or changes without departing from the principles of this application, and can also combine the above technical features in an appropriate manner. These improvements, modifications, changes, or combinations, or the direct application of the inventive concept and technical solution to other situations without modification, should all be considered within the scope of protection of this application.

Claims

1. A method for predicting the lifespan of small household appliance motors based on data analysis, characterized in that, The small household appliance motor utilizes only the existing current sampling circuit in its motor drive circuit, without adding any vibration, temperature, or dedicated sensing devices, and includes the following steps: Step S1: Obtain the working current timing signal output by the current sampling loop, and simultaneously obtain the control signal of the motor drive circuit; Step S2: Preprocess and extract features from the working current timing signal to obtain a set of multidimensional features, including a first subset of features for characterizing the mechanical state and a second subset of features for characterizing the thermal state; Step S3: Input the multidimensional features into the pre-trained health degradation assessment model to obtain the health indicators of small household appliance motors; Step S4: Based on the historical sequence of health indicators, estimate the remaining service life of the small household appliance motor using a life prediction model.

2. The method for predicting the lifespan of small household appliance motors based on data analysis according to claim 1, characterized in that, Step S1 includes: S11: Acquire the operating current timing signal in a manner that is synchronized with the edge trigger of the pulse width modulation signal, wherein sampling is performed at the midpoint of the on or off period in each carrier cycle, and the sampling frequency is not less than 20 times the current electrical frequency. S12: Synchronously acquire the control signal of the motor drive circuit, and reconstruct the electrical angle and load condition information of the motor in real time based on the control signal. The control signal includes a pulse width modulation duty cycle signal, and the electrical angle is used to demodulate the working current timing signal in sync with the mechanical rotation.

3. The method for predicting the lifespan of small household appliance motors based on data analysis according to claim 1, characterized in that, In step S2, the first subset of features used to characterize the mechanical state is obtained through the following steps: The working current timing signal is synchronously demodulated to separate the current ripple component that is synchronized with the mechanical rotation frequency of the small household appliance motor rotor. The current ripple component includes sideband components centered on the electrical frequency and its harmonics, with the offset being the mechanical rotation frequency. The current ripple component is subjected to order tracking analysis to calculate the amplitude of its fundamental frequency integer multiple harmonics. The order tracking analysis adopts an order tracking algorithm based on the Volk-Kalman filter. Select the amplitudes of two target harmonics from the results of the order tracking analysis, and calculate the amplitude ratio between different target harmonics; The selected amplitude and the calculated ratio are combined to form the first type of feature subset.

4. The method for predicting the lifespan of small household appliance motors based on data analysis according to claim 3, characterized in that, The synchronous demodulation process is implemented using orthogonal demodulation technology and includes the following steps: Based on the pulse width modulation duty cycle signal, the real-time electrical angle signal of the motor is reconstructed as a demodulation reference. Using the electrical angle signal as the phase reference, the working current timing signal is decomposed into synchronous and quadrature components; The synchronization component and the quadrature component are combined into a complex current ripple signal, and the instantaneous amplitude and instantaneous phase of the current ripple signal are calculated from the synchronization component and the quadrature component, respectively.

5. The method for predicting the lifespan of small household appliance motors based on data analysis according to claim 1, characterized in that, In step S2, the second subset of features used to characterize the thermal state is obtained through the following steps: Based on the thermal parameters obtained through the calibration of small household appliance motor models, a first-order equivalent thermal network model of the winding is established. Calculate the thermal power of the input model, which is the superposition of Joule thermal power and switching loss power. The Joule thermal power is determined by the effective value of the operating current and the winding resistance, and the switching loss power is determined by the pulse width modulation duty cycle signal and the preset switching loss coefficient. The thermal power is input into the first-order equivalent thermal network model, and the estimated temperature rise curve of the winding is obtained in real time. Features for characterizing fatigue accumulation of insulating materials under thermal cycling stress are extracted from the temperature rise curves, and the extracted features are used as a second subset of features, including thermal inertia features, model mismatch features, and thermal fluctuation features.

6. The method for predicting the lifespan of small household appliance motors based on data analysis according to claim 5, characterized in that: The thermal inertia characteristic is the time it takes for the estimated temperature rise of the winding to reach 63.2% of its steady-state value from the initial value, which is the time constant of the first-order equivalent thermal network model. The model mismatch feature is the coefficient of determination or root mean square error of the fit between the temperature rise curve data points obtained by real-time solution and the theoretical curve generated based on the time constant and the first-order step response model. The thermal fluctuation characteristic is the difference between the peak and trough values ​​of the estimated temperature rise of the winding during a complete and typical working cycle of a small household appliance motor, i.e., the peak-to-peak temperature rise.

7. The method for predicting the lifespan of small household appliance motors based on data analysis according to claim 1, characterized in that, Step S3 includes: S31: Input the first type of feature subset into the first coding network for processing to obtain the mechanical degradation latent variable; S32: Input the second type of feature subset into the second coding network for processing to obtain thermal degradation latent variables; S33: Obtain a query signal reflecting the current operating condition of the motor, the query signal including the effective value of the current operating current, the pulse width modulation duty cycle and the real-time speed reconstructed based on the control signal; S34: Based on the query signal, the fusion weights of the mechanical degradation latent variable and the thermal degradation latent variable are dynamically calculated through the attention fusion module, and weighted fusion is performed to obtain a comprehensive characterization; S35: Map the comprehensive representation to a one-dimensional health indicator.

8. The method for predicting the lifespan of small household appliance motors based on data analysis according to claim 7, characterized in that, The first and second encoding networks are obtained through cross-domain transfer learning, including: A basic feature extraction network is pre-trained on a public multi-sensor dataset containing various types of industrial motors, the multi-sensor data including vibration, temperature and current signals; The parameters of the basic feature extraction network are fixed, and its original input and output layers related to vibration and temperature sensors are removed. An adaptation layer is constructed to map the output dimension of the basic feature extraction network to the single current feature space of the small household appliance motor; Using a dataset of small household appliance motors containing single current signals, the adaptation layer and subsequent regression layer are jointly fine-tuned to obtain a health degradation assessment model.

9. The method for predicting the lifespan of small household appliance motors based on data analysis according to claim 1, characterized in that, Step S4 includes: S41: Based on the start-stop status of small household appliance motors, filter health indicators under stable operating conditions and construct historical sequences for prediction; S42: Input the historical sequence into the exponential degradation model and the lightweight time-series neural network simultaneously to obtain the first lifetime estimate and its goodness of fit output by the exponential degradation model, and the second lifetime estimate and its prediction uncertainty output by the time-series neural network. S43: Based on the goodness of fit and prediction uncertainty, select the first lifetime estimate and the second lifetime estimate, and output the final remaining lifetime estimate.

10. The method for predicting the lifespan of small household appliance motors based on data analysis according to claim 9, characterized in that, The step of selecting between the first lifetime estimate and the second lifetime estimate includes: Compare the goodness of fit with a preset confidence threshold, and the prediction uncertainty with a preset uncertainty threshold; If the goodness of fit is higher than its threshold and the prediction uncertainty is lower than its threshold, it indicates that the two models predict in a consistent and reliable manner, and the weighted average of the two models is output as the final remaining lifetime. If the goodness of fit is lower than its threshold or the prediction uncertainty is higher than its threshold, it indicates that the current prediction uncertainty is high. The conservative warning mode is activated, and the shorter lifespan value predicted in the dual model is output as the remaining lifespan estimate. An auxiliary prompt message "low prediction reliability, it is recommended to shorten the inspection cycle" is generated.

Images