A screw tightening abnormal data detection method based on unsupervised learning

Abstract

The application discloses a screw tightening abnormal data detection method based on unsupervised learning and belongs to the technical field of industrial time series data anomaly detection. Based on related technologies in the field of deep learning, the application completes unsupervised anomaly detection, extracts data features through a recurrent neural network and a time convolution network, more fully learns time correlation features of screw tightening data, combines an autoencoder to perform dimension reduction and reconstruction, realizes unsupervised missed detection data anomaly detection, and thus realizes more efficient and accurate anomaly curve detection without human participation and realizes identification of screw tightening abnormal data.

Description

Technical Field

[0001] This invention relates to a method for detecting abnormal screw tightening data based on unsupervised learning, belonging to the field of industrial time-series data anomaly detection technology. Background Technology

[0002] Screw tightening equipment, as one of the most commonly used parts in industrial production, directly impacts the quality of industrial products. Taking the automotive manufacturing industry as an example, there are approximately thousands of screw connections during vehicle assembly. A significant portion of these connections are critical, controlling the tightening quality of key automotive components. Problems during tightening can lead to serious safety accidents. Therefore, screw tightening anomaly detection technology is widely used in industrial manufacturing.

[0003] Traditional screw tightening anomaly detection technologies only perform simple analysis of industrial time-series data, failing to identify key characteristics between these data points. For example, the commonly used final torque determination method judges the tightening process as normal simply by checking if the final torque falls within a predetermined range, making it impossible to identify some anomalies, compromising product quality, and creating potential safety hazards. Furthermore, the sheer volume of screw curve data, with normal data far outweighing anomalies, makes anomaly detection difficult to perform manually and with equipment alone. Summary of the Invention

[0004] The purpose of this invention is to address the defects and shortcomings of existing technologies, and to creatively propose an unsupervised method for detecting abnormal data in screw tightening data that has been missed by equipment without human intervention. This method can achieve accurate and efficient anomaly detection.

[0005] The screw tightening process is mainly divided into three stages: the free tightening stage before sufficient contact, the torque increase stage after initial contact, and the tightening stage within the elastic deformation range. The curve data samples plotted from the recorded torque values ​​throughout this process exhibit strong temporal correlation characteristics. Therefore, this invention leverages the well-established recurrent neural networks and temporal convolutional networks in deep learning time series processing to fully extract the temporal correlation features of the curve data. Subsequently, an autoencoder network structure is used for feature compression and reconstruction. Abnormal data is detected by analyzing the relationship between the reconstruction error and a threshold.

[0006] Recurrent Neural Networks (RNNs) are a type of recursive neural network that takes sequence data as input, recurses along the direction of sequence evolution, and connects all nodes (recurrent units) in a chain-like manner. RNNs possess memory, parameter sharing, and Turing completeness, thus offering advantages in learning the nonlinear features of sequences. RNNs are applied in Natural Language Processing (NLP), such as speech recognition, language modeling, and machine translation, and are also used for various time series data processing. RNNs constructed using Convolutional Neural Networks (CNNs) can handle computer vision problems involving sequential inputs.

[0007] Long Short-Term Memory (LSTM) is a special type of RNN primarily designed to address the vanishing and exploding gradient problems during the training of long sequences. Internally, it uses gating units to selectively retain past memories, enabling it to learn the most important features of long sequences and thus perform better on even longer sequences. Bidirectional LSTM is an improved version of LSTM, learning features simultaneously from two directions, thus overcoming the limitation of unidirectional LSTM in effectively connecting to context.

[0008] Temporal Convolutional Networks (TCNs) are an improvement on convolutional neural networks. By incorporating causal convolutions and dilated convolutions, they gain the ability to process time-series data. Furthermore, through residual connections between networks, they avoid the vanishing or exploding gradient phenomena. In addition, TCNs retain the parallelization capabilities of convolutional networks and do not suffer from the limitations of inter-network connections inherent in recurrent neural networks, making their training more efficient.

[0009] An autoencoder (AE) is a widely used neural network model in semi-supervised and unsupervised learning. It learns the latent features of input data (encoding) and reconstructs the original input data using these learned features (decoding). Intuitively, autoencoders can be used for feature reduction, similar to Principal Component Analysis (PCA), but they are more efficient because neural network models can extract more effective new features. Besides feature reduction, the new features learned by an autoencoder can be fed into supervised learning models; therefore, autoencoders can act as feature extractors.

[0010] The innovation of this invention lies in the following: Based on deep learning-related technologies, unsupervised anomaly detection is achieved. Data feature extraction is performed through recurrent neural networks and temporal convolutional networks to more fully learn the temporal correlation features of screw tightening data. Combined with an autoencoder for dimensionality reduction and reconstruction, unsupervised detection of missed data anomalies is achieved. This enables more efficient and accurate anomaly curve detection without human intervention, thus enabling the identification of screw tightening anomaly data.

[0011] The present invention is achieved using the following technical solution.

[0012] A method for detecting abnormal screw tightening data based on unsupervised learning includes the following steps:

[0013] Step 1: Initialize system parameters. This includes neural network model parameters, learning rate, data batch size, number of iterations, and error threshold.

[0014] Step 2: Preprocess and normalize the data. Preprocess all sample data to a uniform length and normalize the data according to the data range.

[0015] Step 3: Divide the sample data into training set, validation set and test set according to the set ratio.

[0016] Step 4: Train the neural network model on the training and validation sets, adjust the model parameters using the backpropagation algorithm, and end the process after reaching the specified number of iterations, then save the model.

[0017] Step 5: Feed the test set into the trained model to perform anomaly detection.

[0018] Beneficial effects

[0019] The method of this invention adopts an unsupervised approach, eliminating the need for manual data annotation and significantly reducing labor costs. Furthermore, the combination of temporal convolutional networks and recurrent neural networks enables the full extraction of key features from the data, achieving more efficient and accurate detection results. Attached Figure Description

[0020] Figure 1 This is a flowchart of the method of the present invention.

[0021] Figure 2 This is a schematic diagram of the neural network model of the method of the present invention; Specific implementation methods

[0022] The method of the present invention will be further described in detail below with reference to the accompanying drawings.

[0023] like Figure 1 As shown, this paper presents a method for detecting abnormal screw tightening data based on unsupervised learning. Its neural network model is an autoencoder structure, where the encoder part is based on a recurrent neural network and a temporal convolutional network, and the decoder part is composed of a two-layer recurrent neural network. Figure 2 As shown.

[0024] Specifically, it includes the following steps:

[0025] Step 1: Preset the parameters of the neural network model.

[0026] Specifically, prior knowledge is used to preset various parameters in the neural network model. Among them, the maximum number of iterations is E, E = 200, the learning rate is α, α = 0.001, and the data batch size is B, B = 64.

[0027] Step 2: Data preprocessing and normalization.

[0028] By sampling, the screw tightening sample data is preprocessed to a uniform length T and then subjected to maximum and minimum normalization.

[0029] Specifically, for sample data S, with length T, maximum value U, and minimum value D, the value of each point is set as follows:

[0030]

[0031] Among them, S i This represents the i-th sample data.

[0032] Step 3: Sample set partitioning.

[0033] The sample dataset is divided into training, validation, and test sets in an 8:1:1 ratio (or other ratios).

[0034] Step 4: Train the neural network model.

[0035] The model is trained on the training and validation sets. During training, the error at each iteration is L. For each model parameter w, its value is updated using the backpropagation algorithm:

[0036]

[0037] Repeat this process until the number of iterations E is reached.

[0038] Step 5: Feed the test set into the trained neural network model to detect screw tightening data anomalies.

[0039] Let the error threshold be θ. For sample data S, if its reconstruction error l after passing through the model satisfies:

[0040] l> θ (3)

[0041] Then the sample data S is considered to be outlier. Example

[0042] This example details the process of an algorithm for detecting abnormal screw tightening data based on unsupervised learning.

[0043] Step 1: First, preset the system parameters.

[0044] Based on prior knowledge, the model training iterations are set to 200, the learning rate to 0.001, and the batch size to 64.

[0045] Step 2: In practice, the time-series data collected in production is used. First, the sample data is uniformly processed into data with a length of 600 using the sampling method. Then, the data is normalized according to Equation 1.

[0046] Step 3: Divide the sample dataset into training, validation, and test sets in an 8:1:1 ratio;

[0047] Step 4: Train the model on the training and validation sets. Adjust the model parameters according to Equation 2 for each training iteration. Stop after reaching the maximum number of iterations of 200 and save the model parameters.

[0048] Step 5: Feed the test set into the trained model and perform anomaly detection according to Equation 3.

[0049] The above description is merely a preferred embodiment of the present invention, and the present invention should not be limited to the content disclosed in this embodiment and the accompanying drawings. Any equivalent or modified embodiments made without departing from the spirit of the present invention fall within the scope of protection of the present invention.

Claims

1. A method for detecting abnormal screw tightening data based on unsupervised learning, characterized in that, Data feature extraction is performed using recurrent neural networks and temporal convolutional networks to learn the temporal correlation features of screw tightening data. This is combined with an autoencoder for dimensionality reduction and reconstruction to achieve unsupervised detection of missed data anomalies. The recurrent neural network model is an autoencoder structure, where the encoder part is based on a recurrent neural network and a temporal convolutional network, and the decoder part is composed of a two-layer recurrent neural network. Step 1: Initialize system parameters, including neural network model parameters, learning rate, data batch size, number of iterations, and error threshold; Step 2: Preprocess and normalize the data. Preprocess all sample data to a uniform length and normalize the data according to the data range. Step 3: Divide the sample data into training set, validation set and test set according to the set ratio; Step 4: Train the neural network model on the training and validation sets, adjust the model parameters using the backpropagation algorithm, and end the process after reaching the specified number of iterations, then save the model. Step 5: Feed the test set into the trained model to perform screw tightening data anomaly detection.

2. The method for detecting abnormal screw tightening data based on unsupervised learning as described in claim 1, characterized in that, In step 2, the screw tightening sample data is preprocessed into a uniform length T by sampling, and then subjected to maximum and minimum normalization. For sample data S, with length T, maximum value U, and minimum value D, set the value of each point as follows: Among them, S i This represents the i-th sample data.

3. The method for detecting abnormal screw tightening data based on unsupervised learning as described in claim 1, characterized in that, In step 3, the training set, validation set, and test set are divided in an 8:1:1 ratio.

4. The method for detecting abnormal screw tightening data based on unsupervised learning as described in claim 1, characterized in that, In step 4, during model training, the error for each iteration is L. For each model parameter w, its value is updated using the backpropagation algorithm: Repeat this process until the number of iterations E is reached.

5. The method for detecting abnormal screw tightening data based on unsupervised learning as described in claim 1, characterized in that, In step 5, the test set is fed into the trained neural network model to detect anomalies in screw tightening data; let the error threshold be θ. For sample data S, if its reconstruction error l after passing through the model satisfies: l>θ Then the sample data S is considered to be outlier.

Images