Search

CN-122020046-A - Method for identifying bacterial contamination of inner cylinder of washing machine based on trace detection

CN122020046ACN 122020046 ACN122020046 ACN 122020046ACN-122020046-A

Abstract

The invention discloses a trace detection-based method for identifying bacterial contamination in an inner cylinder of a washing machine, which relates to the technical field of intelligent household appliance monitoring and comprises the following steps of S1, constructing a four-dimensional state point sequence, S2, constructing a Gaussian mixture model fingerprint library, S3, outputting a topological metric value, S4, inputting the four-dimensional state point sequence into an improved FlowFormer model, analyzing nonlinear dynamics characteristics of bacterial growth based on regularized probability distribution transformation of an optimal transmission path, extracting a track tangent space base vector, S5, extracting a main curvature direction and an evolution rate parameter of a nonlinear track of a manifold subspace in a space-time dimension, S6, outputting an updated Gaussian mixture model fingerprint library and a corresponding translation compensation vector, and S7, matching relative topological positions. The invention overcomes the limitations of low signal-to-noise ratio, difficult biological activity detection and ignoring evolution trend of the traditional method, and provides a high-efficiency solution for the accurate sterilization of the intelligent washing machine.

Inventors

  • XIAO XUHUA

Assignees

  • 广东嘉富数字科技有限公司

Dates

Publication Date
20260512
Application Date
20260130

Claims (8)

  1. 1. The method for identifying bacterial contamination of the inner cylinder of the washing machine based on trace detection is characterized by comprising the following steps of: S1, collecting sensor data, rotating speed and phase angle of a washing machine, solving energy parameters of a flow field by combining a fabric porous medium model, carrying out convolution weighting by a time-varying filter, and executing in-phase difference based on the phase angle to construct a four-dimensional state point sequence; S2, processing historical data based on HDBSCAN algorithm, modeling a bacterial biomembrane zone into a special-shaped high-density cloud cluster, constructing a Gaussian mixture model fingerprint library, and fitting a clean zone, a physical dirt zone and the bacterial biomembrane zone; S3, constructing a covariance matrix by utilizing a Gaussian mixture model fingerprint library, calculating the mahalanobis distance between each state point in the four-dimensional state point sequence and each high-density cloud cluster centroid, and outputting a topology metric value; S4, inputting the four-dimensional state point sequence into an improved FlowFormer model, analyzing nonlinear dynamics characteristics of bacterial growth based on regularized probability distribution transformation of an optimal transmission path, and extracting a track tangent space base vector of a pollution evolution trend in a manifold subspace; S5, carrying out logarithmic mapping on the trajectory tangent space base vector by the base Yu Liman manifold geometry, straightening a nonlinear trajectory of the manifold subspace to a tangent plane of the Euclidean space, and extracting a principal curvature direction and an evolution rate parameter in a space-time dimension through a tensor voting mechanism; S6, tracking cloud mass center offset of the clean zone by utilizing the self-adaptive sliding window, performing coordinate dynamic translation transformation on the Gaussian mixture model fingerprint library, and outputting an updated Gaussian mixture model fingerprint library and a corresponding translation compensation vector; And S7, carrying out coordinate correction on the topological metric value, the main curvature direction and the evolution rate parameter based on the translation compensation vector, combining the updated Gaussian mixture model fingerprint library to match the relative topological position, triggering a powerful degerming signal if the central area is the central area, and outputting an alarm signal if the central area is the boundary area and the track trend points to the inside.
  2. 2. The method for identifying bacterial contamination in an inner cylinder of a washing machine based on trace detection according to claim 1, wherein the step S1 specifically comprises: S11, based on collected sensor data of the washing machine and read rotating speed and phase angle of a roller, constructing a three-dimensional unstructured grid of a fluid field in the roller, inputting the rotating speed and the phase angle of the roller as dynamic boundary conditions into a Navier-Stokes equation, simulating a clothing permeation effect by combining a fabric porous medium model, solving a speed field and a pressure field of the fluid field through a finite element analysis method, and calculating real-time energy distribution parameters applied to a sensor sensing area by the fluid field; S12, constructing a time-varying filter bank by utilizing real-time energy distribution parameters of a flow field, carrying out time-frequency domain weighting treatment on sensor data of the washing machine through convolution operation, filtering non-fluid load interference components, and generating weighted characteristic data; And S13, performing periodical slice and phase alignment on the weighted characteristic data based on the phase angle of the roller, calculating characteristic difference values of adjacent aligned slices at the same phase moment, and constructing a four-dimensional state point sequence containing position coordinates and motion vectors.
  3. 3. The method for identifying bacterial contamination in an inner cylinder of a washing machine based on trace detection according to claim 1, wherein the step S2 specifically comprises: S21, analyzing the spatial distribution density and the topological connectivity of the data samples by utilizing HDBSCAN algorithm based on collected historical sensor data of the washing machine, aggregating the discrete data samples into special-shaped high-density cloud clusters, and modeling the bacterial biofilm area into the special-shaped high-density cloud clusters; S22, constructing a Gaussian mixture model based on morphological characteristics of the special-shaped high-density cloud cluster, and fitting probability density distribution of historical data through the Gaussian mixture model to generate a Gaussian mixture model fingerprint library comprising characteristics of a clean area, a physical dirt area and a bacterial biofilm area; s23, performing multidimensional parameter fitting on the real-time sensor data by utilizing a Gaussian mixture model fingerprint library, dividing the space boundaries of the clean area, the physical dirt area and the bacterial biofilm area according to the membership probability of fitting, and outputting real-time state distribution parameters of each area.
  4. 4. The method for identifying bacterial contamination in an inner cylinder of a washing machine based on trace detection according to claim 1, wherein the step S3 specifically comprises: s31, based on a Gaussian mixture model fingerprint library, gaussian component parameters corresponding to a clean area, a physical dirt area and a bacterial biofilm area are extracted, and an inverse covariance weighting matrix is constructed by utilizing covariance matrixes of all Gaussian components; s32, reading state point coordinate vectors in the four-dimensional state point sequence, substituting an inverse covariance weighting matrix into a Markov distance measure formula, and calculating the weighted distance from each state point coordinate vector to the mass center of the special-shaped high-density cloud cluster; s33, mapping the weighted distance normalization into a topology metric value, judging the space attribution attribute of the state point according to the size of the topology metric value, binding the judging result with the state point, and outputting a topology metric sequence.
  5. 5. The method for identifying bacterial contamination in an inner drum of a washing machine based on trace detection according to claim 1, wherein the improved FlowFormer model comprises a spatiotemporal feature embedding layer, a flow vector field prediction layer, a flow matching integration layer and an evolution trend coding layer: The space-time feature embedding layer is used for receiving a four-dimensional state point sequence, extracting a long-range space-time dependency relationship between state points by using a self-attention mechanism of a transducer encoder, mapping an input sequence to initial feature distribution of a high-dimensional potential space, and outputting the initial feature distribution; The flow vector field prediction layer is used for receiving initial characteristic distribution and current time step conditions, performing time step embedded coding on the current time step conditions to obtain time condition vectors, fusing the time condition vectors and the initial characteristic distribution in characteristic dimensions to obtain fused space-time context characteristics, inputting the fused space-time context characteristics into a multi-layer perceptron to perform nonlinear transformation, fitting a conditional speed field function defined by a flow matching model based on a transducer architecture, and calculating a speed vector for driving a current state point to evolve towards a target probability density distribution to be used as a vector field for transferring from a current moment state to a next moment state; The flow vector field integration layer is used for carrying out weighted summation by utilizing the state characteristics of the current moment and the speed vector based on a preset time step, executing forward Euler integration update to acquire the state characteristics of the next moment, carrying out recursion iteration on the forward Euler integration update process along the time dimension, constructing a continuous evolution track of a state point under the drive of a vector field, generating a track evolution path formed by the state characteristics of the continuous moment, calculating the mean square error between the speed vector and the real tangent vector of the preset optimal transmission path, taking the mean square error as a flow matching loss, updating the parameters of the flow vector field prediction layer by a backward propagation algorithm, and optimizing the prediction precision of a vector field; The evolution trend coding layer is used for receiving the track evolution path, carrying out coding analysis on the track evolution path, extracting a tangential space direction vector and curvature change parameters of the track in the manifold latent space, and outputting a track tangential space base vector of the pollution evolution trend.
  6. 6. The method for identifying bacterial contamination in an inner cylinder of a washing machine based on trace detection according to claim 1, wherein the step S5 specifically comprises: S51, calculating a Riemann metric matrix of a track at a current position point of a manifold latent space based on a Riemann manifold geometric theory, and carrying out Cholesky decomposition on the Riemann metric matrix to obtain a linear transformation matrix; mapping the manifold tangential space base vector to a tangential plane coordinate system by utilizing a linear transformation matrix, performing nonlinear transformation from an exponential coordinate to a logarithmic coordinate, conformally projecting a nonlinear track structure of a manifold subspace to a tangential plane of an Euclidean space, and generating a linear feature vector representing a track evolution trend; s52, in a tangential plane of Euclidean space, encoding the linear feature vector into an initial second-order symmetrical tensor, and introducing attenuation weight based on a Gaussian kernel function to construct a tensor voting field so as to propagate structural information of a neighborhood point to a current voting center; S53, carrying out feature decomposition on the accumulated tensor voting field, judging the significance of the track according to whether the difference value between the maximum feature value and the next-largest feature value is larger than a preset threshold value, determining the main curvature direction in the space-time dimension by utilizing the feature vector direction corresponding to the maximum feature value, and calculating the evolution rate parameter based on the modular length of the maximum feature value.
  7. 7. The method for identifying bacterial contamination in an inner cylinder of a washing machine based on trace detection according to claim 1, wherein the step S6 specifically comprises: s61, setting the time length of a self-adaptive sliding window based on the observation data of the current moment of the clean zone, and carrying out weighted barycenter calculation on pixel coordinates of the observation data at each moment in the self-adaptive sliding window to obtain weighted barycenter coordinates of the current moment; S62, based on a Gaussian mixture model fingerprint library, analyzing distribution parameters of all Gaussian components in the Gaussian mixture model, constructing a global translation matrix by utilizing cloud mass center offset, applying the global translation matrix to mean vector parameters of all Gaussian components, and performing matrix addition operation to update the mean vector parameters to generate an updated Gaussian mixture model fingerprint library; And S63, extracting translation components in the global translation matrix as translation compensation vectors, and outputting the updated Gaussian mixture model fingerprint library and the corresponding translation compensation vectors.
  8. 8. The method for identifying bacterial contamination in an inner cylinder of a washing machine based on trace detection according to claim 1, wherein the step S7 specifically comprises: S71, based on a translation compensation vector, obtaining an original coordinate of a target observation point under an initial physical coordinate system, performing matrix translation transformation on the original coordinate, and calculating to obtain a corrected coordinate of the target observation point under a reference coordinate system; S72, calculating the Mahalanobis distance of each Gaussian component in the updated Gaussian mixture model fingerprint library by using the corrected coordinates, updating the value of the Mahalanobis distance into a corrected topological metric value, selecting the Gaussian component corresponding to the minimum Mahalanobis distance as the optimal matching component, and determining the relative topological position; S73, analyzing a spatial attribute label corresponding to the optimal matching component, judging whether the optimal matching component belongs to a bacterial biomembrane zone core set, triggering a powerful degerming signal if the optimal matching component belongs to the bacterial biomembrane zone core set, judging whether the optimal matching component belongs to a topological junction set if the optimal matching component does not belong to the bacterial biomembrane zone core set, and calculating whether a track trend points to the inside of a bacterial biomembrane zone by combining the corrected evolution rate parameter and the corrected track tangential space direction if the optimal matching component belongs to the topological junction set, and outputting an alarm signal if the track trend points to the inside of the bacterial biomembrane zone.

Description

Method for identifying bacterial contamination of inner cylinder of washing machine based on trace detection Technical Field The invention relates to the technical field of intelligent household appliance monitoring, in particular to a method for identifying bacterial contamination of an inner cylinder of a washing machine based on trace detection. Background With the wide deployment of sensing nodes of the internet of things of intelligent household appliances, the scale of the running state monitoring data of the inner barrel of the washing machine presents an exponential rising trend, and the traditional sensing signal processing method faces serious signal-to-noise ratio challenges in the aspects of feature extraction and early warning of trace pollution. The existing stain recognition algorithm based on threshold judgment, such as a support vector machine and a shallow neural network, improves the classification efficiency by utilizing a statistical principle, but mainly relies on the mean or variance of the energy spectrum of the data acquired by the sensor to carry out static feature modeling. The method based on the statistical features omits the hidden complicated nonlinear dynamics mechanism (such as topological continuity of probability distribution transformation and geometric constraint of an optimal transmission path) and deep manifold space-time evolution association in the bacterial biofilm growth process in a flow field environment, so that the structural distortion of a high-dimensional feature space occurs when a polluted fingerprint library is constructed, and the early recognition accuracy for micro-biofilm pollution is limited. In addition, classical stain identification methods often have difficulty in fully utilizing the Riemann manifold geometric features in multi-dimensional sensor data to constrain the evolution trend prediction space, resulting in often large numbers of false positives or false negatives when dealing with dynamic tracking of bacterial growth and diffusion trends, increasing the computational overhead of deep cleaning decisions. Therefore, how to provide a method for identifying bacterial contamination in an inner cylinder of a washing machine based on trace detection is a problem to be solved by those skilled in the art. Disclosure of Invention The invention provides a trace detection-based method for identifying bacterial contamination in an inner cylinder of a washing machine, which is characterized in that a nonlinear dynamics analysis mechanism based on an optimal transmission path is established by introducing an improved FlowFormer model, and manifold latent space mapping and space-time dependency extraction are respectively carried out on a four-dimensional state point sequence and a Gaussian mixture model fingerprint library. The improved FlowFormer model analyzes nonlinear dynamics characteristics of bacterial growth through regularized probability distribution transformation of an optimal transmission path, and extracts high-dimensional manifold characteristics comprising principal curvature directions and trajectory tangent space basis vectors. And carrying out space matching on the multidimensional feature vector corrected based on the translation compensation vector and the updated Gaussian mixture model fingerprint library, and judging the bacterial growth trend by analyzing the dot product relation between the track tangential vector and the normal vector of the biological film region. According to the mechanism, by establishing a self-adaptive feedback path from the improved FlowFormer model evolution track prediction to the dynamic fingerprint library coordinate translation, the positioning deviation caused by the cloud mass center drift of the clean area is effectively eliminated, the generated bacterial pollution identification result is ensured to dynamically keep the nonlinear topological characteristic of bacterial biofilm growth, and the technical effects of accurately identifying trace bacterial pollution and early warning the diffusion trend under the interference of a complex flow field are realized. The invention overcomes the limitations of low signal-to-noise ratio, difficult biological activity detection and ignoring evolution trend of the traditional method, and provides a high-efficiency solution for the accurate sterilization of the intelligent washing machine. The method for identifying bacterial contamination of the inner cylinder of the washing machine based on trace detection, provided by the embodiment of the invention, comprises the following steps: S1, collecting sensor data, rotating speed and phase angle of a washing machine, solving energy parameters of a flow field by combining a fabric porous medium model, carrying out convolution weighting by a time-varying filter, and executing in-phase difference based on the phase angle to construct a four-dimensional state point sequence; S2, processing historical data based on HD