Tower crane tower body damage abnormity detection method

Abstract

The invention provides a tower crane tower body damage anomaly detection method, and relates to the field of anomaly detection. Comprising a selection fragment mapping module, a phase period anomaly modeling module and an anomaly judgment module, the selection fragment mapping module is used for local disturbance analysis, the phase period anomaly modeling module is used for processing periodic anomaly information, and the anomaly judgment module is used for integrating two scores to obtain a final anomaly score and complete anomaly judgment. All the modules are matched with one another to achieve abnormal detection of damage to the tower body of the tower crane.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of anomaly detection, and particularly relates to a tower crane tower damage anomaly detection method. BACKGROUND

[0002] As the core hoisting equipment in high-rise building construction, tower cranes are widely used in large-scale construction, industrial facilities and other engineering projects. The tower structure is usually composed of multiple steel structures, which are gradually increased in height by standard sections. It has the characteristics of high height, heavy load and complex structure. Due to its long-term high load, frequent vibration and wind environment, the tower structure is prone to fatigue damage, loose connection and other problems. If the damage is not discovered and handled in time, it is easy to cause structural instability, hoisting accidents and even overall collapse, causing significant property loss and casualties.

[0003] Current structural health detection of tower cranes mainly relies on manual inspection and vibration monitoring. These methods have problems such as long detection period, low efficiency and strong subjectivity, making it difficult to discover early and small damage in time. At the same time, the results of manual inspection are often limited by the experience of the inspector and the coverage of the equipment. Although vibration monitoring can effectively identify the dynamic changes of the tower, it is difficult to effectively extract features with high discriminability from a large amount of vibration data.

[0004] In recent years, with the continuous development of intelligent and data-driven technologies, damage detection methods based on time series data analysis have gradually become a new research trend. These methods can more accurately identify the small damage of the tower by deeply analyzing the vibration, tilt, load and other time series data collected during the operation of the tower. Tower damage usually manifests as local deformation, crack propagation, vibration frequency change and other abnormal behaviors. These abnormalities are accompanied by sudden nonlinear dynamic fluctuations or periodic disturbances, and have the characteristics of multi-scale and multi-dimensional on data. By deeply analyzing these data, potential risks of the tower can be discovered in time. SUMMARY

[0005] The application provides a tower crane tower damage anomaly detection method. For local nonlinear, sudden and periodic tower crane tower data, a CTS-DAM anomaly detection model is proposed, which includes a selected segment mapping module, a phase cycle anomaly modeling module and an anomaly judgment module.

[0006] The technical solution adopted by the application to achieve the above purpose specifically includes the following steps: S1, collect tower crane tower related data, construct a data set, and divide the training set, validation set and test set after preprocessing; S2, construct a selected segment mapping module for local disturbance analysis, the specific steps are as follows: S21, the input tower crane tower data is divided into multiple scales, a feature encoding method of introducing self-attention mechanism is introduced, and tower crane tower feature embedding is calculated; S22, the tower crane tower feature embedding is calculated to calculate the structure disturbance factor, and a structure disturbance tensor representation matrix is constructed to obtain the structure selective gating tensor; S23, based on the relationship between the structure disturbance factor and the structure prototype vector, a plurality of coupled driving global scaling factors are designed; S24, the output of S21 and S22 and the plurality of coupled driving global scaling factors are introduced to calculate the structure selective mapping weight, reconstruct the restored tower crane tower sequence, and obtain the selection slice mapping score; S3, a phase response driving module is constructed for processing periodic abnormal information, and the specific steps are as follows: S31, input the restored tower crane tower sequence, project to obtain the phase trajectory tensor, perform sliding window operation and construct the phase evolution induction map; S32, define a prototype graph set, calculate the structural deviation metric between the phase evolution induction map and the prototype graph, and design a gating factor to dynamically adjust the structural expression of the prototype graph through the gating mechanism; S33, by calculating the distance relationship between the induction map and a plurality of updated prototype graphs, the phase response deviation score is obtained; S4, an abnormality judgment module is constructed, the phase response deviation score and the selection slice mapping score are weighted and fused, and a threshold is set to judge the damage abnormality of the tower crane tower.

[0007] Preferably, in S1, the tower crane tower related data is collected, including tower vibration sensor data, tower stress sensor data, tower temperature sensor data, tower displacement sensor data and tower ultrasonic wave detection data, a multi-dimensional data set is constructed, data preprocessing is performed using mean filling method, and the training set, the validation set and the test set are divided according to the ratio of 7:1:2.

[0008] Preferably, the tower crane tower is easily affected by load change, wind vibration disturbance and environmental aging in long-term operation, and presents significant multi-scale non-stationary structural change characteristics. Traditional methods mostly use single-scale modeling, which is difficult to capture weak damage characteristics at each scale, especially in the early abnormal stage, which is often ignored due to unobvious structure response. In order to realize more accurate structure state identification, it is necessary to introduce multi-scale analysis and selective modeling mechanism, so as to efficiently encode the embedded representation of structure fragments at different scales.

[0009] Preferably, in S21, the tower crane tower data is input by setting fixed scale parameters The input tower crane tower data is subjected to non-overlapping mean division processing at each scale, and the specific mathematical model is: ; In the formula, is a scale parameter, indicating the slice scale, is the number of the i-th slice, is the end time step number of the slice, is the tower crane tower data of the t-th time step, is the i-th slice of the tower crane tower after averaging under the size z, is the mean operation, and the feature encoding method introducing self-attention mechanism is used to calculate the tower crane tower feature embedding The tower crane tower slice vector is mapped to query , key and value by three independent linear transformations, , , wherein , , are learnable weight matrices, and then the self-attention calculates the similarity score through the matching of the query and the key. The specific mathematical model is: ; In the formula, is an activation function, is the feature embedding dimension, is the transpose of the key, and the feature embedding of each scale is represented as: ; In the formula, is the similarity score of the i-th query to the j-th key, is the j-th value, is the number of tower crane tower slices under the scale z, is the slice index being noticed, indicating the position of all keys and values.

[0010] Preferably, a fixed scale division and self-attention mechanism combined modeling method is introduced to embed the tower structure fragments under different scales, establish the attention weight between the structure fragments through the matching of the query and the key, realize the saliency enhancement of the key structure region, improve the sensitivity of the model to the structure disturbance, enhance the discriminability of the feature expression, provide a multi-scale unified feature basis for subsequent structure disturbance response modeling and cycle anomaly detection, and break through the limitations of traditional methods in scale adaptability and structure perception ability.

[0011] Preferably, the operating state of the tower crane tower structure usually has high complexity and local difference, and the traditional feature mapping method often ignores the influence of structure disturbance on the prototype mapping process, making it difficult to accurately depict the structural deviation characteristics of abnormal fragments. In addition, there are significant response differences between different structural fragments. If all structures are directly modeled uniformly, it is easy to lead to insufficient feature generalization or response ambiguity. Therefore, a structure disturbance perception mechanism needs to be introduced in the modeling process to enhance the model's ability to distinguish local differences in structures and achieve selective control of structure mapping relationships.

[0012] Preferably, in the S22, the tower crane tower feature embedding is input , a disturbance response strategy is proposed to reduce all tower crane tower feature embeddings at the current scale to obtain a reference center representation at scale z , and the specific mathematical expression is: ; In the formula, is the number of tower crane tower slices at scale z, is the index of the slice, and the structural difference of each scale of the tower crane tower feature embedding relative to the reference center representation is calculated to form a measurable structural disturbance factor, and the specific mathematical expression is: ; In the formula, is the structural disturbance factor of the i-th tower crane tower feature embedding at scale z, is the Euclidean distance square, and then based on the response offset relationship between different tower crane tower feature embeddings in the embedding space, a structural disturbance tensor representation matrix is constructed, and the specific mathematical expression is: ; In the formula, is the structural disturbance tensor representation matrix, is the softmax normalization of the matching distribution of each slice at each scale z, is the j-th structural prototype vector at scale z, representing a predefined structural template, is a nonlinear mapping function, is the number of structural prototype vectors, is the index of the structural disturbance factor, is the structural prototype vector index, by jointly considering the tower crane tower feature embedding of the current slice, the structural prototype vector and the structural disturbance factor, and combining the trainable linear discriminant parameter, a logical judgment is made on whether to activate each structural prototype path, and a structural selective gating tensor is obtained, and the specific mathematical expression is: {G}^{z}_{ij}=\sigma ({W}_{g}\cdot \left [ {{q}^{z}_{i}||{b}^{z}_{j}||{\Delta}^{z}_{i}} \right ]+{b}_{g}) ; wherein, is an activation function, is a gating weight matrix, is a gating bias term, is a structure-selective gating tensor.

[0013] Preferably, a structure-selective gating mechanism based on the structure disturbance tensor is introduced, a disturbance-aware control factor is embedded in the mapping path between the original structure segment and the prototype, and the gating expression is constructed by jointly considering the structure disturbance factor, the structure prototype vector and the segment embedding representation, so as to realize the dynamic selection and nonlinear suppression of each structure path, ensure that the key structure information is effectively preserved and the disturbance region is highlighted, which not only enhances the sensitivity of the model to local changes in the structure, but also realizes the construction of non-normalized mapping weights based on disturbance adjustment, effectively avoiding the problem of significant loss under normalization operation in the traditional attention mechanism, and improves the modeling ability of abnormal structure response.

[0014] Preferably, the traditional method directly performs simple average or weighted summation on the mapping result, ignoring the matching relationship and disturbance sensitivity between different structure segments, which is difficult to accurately reflect the overall deviation degree of the structure. At the same time, the tower crane tower has high structural complexity and significant local differences, and a feature fusion mechanism considering the matching accuracy of segments and the adjustment of disturbance degree needs to be designed to support more robust abnormal perception criterion construction.

[0015] Preferably, in the S23, a plurality of coupled driving global scaling factors are designed, a nonlinear enhancement term and a periodic modulation factor are fused and constructed based on the relationship between the structure disturbance factor and the structure prototype vector, and the scaling factor corresponding to each slice is obtained through coupled modeling, and the specific mathematical model is: ; wherein, is a plurality of coupled driving global scaling factors of the i-th slice at scale z, is an amplification factor of the polynomial modulation term, is a structure disturbance factor of the i-th tower crane tower feature embedding at scale z, is a polynomial function of the i-th slice structure disturbance factor, is a scaling factor of the control period term frequency, is a frequency regulation index of the i-th slice, a coefficient for controlling phase shift, a structure prototype vector at scale z, an average similarity with the current query, a power index for controlling the amplitude strength of the period.

[0016] Preferably, by fusing the structure selective gating weight, the prototype matching similarity and the disturbance penalty factor, a non-normalized structure mapping fusion mechanism is constructed, on the basis of obtaining the structure path mapping results of each segment, an inhibitory adjustment term is introduced to weight and inhibit the structure segments with high disturbance intensity, so as to avoid the interference of high disturbance area on the overall structure score, and finally output the structure deviation score as the abnormality perception basis. This method breaks the normalization constraint in the traditional mapping fusion process, improves the accurate characterization ability of local anomaly of complex structure, and provides a highly consistent structure expression basis for subsequent periodic anomaly modeling.

[0017] Preferably, the damage evolution of the tower structure of the tower crane often has time sequence characteristics such as periodicity, volatility and local asynchronicity. The traditional method based on statistical analysis or time domain trend detection is difficult to fully reveal the dynamic correlation characteristics of the structure state between adjacent periods. At the same time, in the periodic data, there is a complex phase coupling relationship between different variables, and only relying on single variable or original time sequence modeling may cause incomplete expression of time sequence characteristics. Therefore, a mechanism with periodic modeling ability and variable phase perception ability needs to be introduced to extract global and local dynamic change rules from the time sequence evolution track, so as to realize accurate perception of periodic anomaly.

[0018] Preferably, in the S24, a structure selective mapping weight generation strategy is proposed to fuse multiple factors, a structure selective mapping weight is constructed, a matching relationship between each slice and a structure prototype vector is scaled and adjusted based on a plurality of coupled driving global scaling factors, a scaled matching strength is weighted in slice specificity combined with a structure selective gating tensor, then a matching similarity between the slice and the structure prototype vector is introduced as a basic structure similarity expression, and each element in a structure disturbance tensor representation matrix is multiplied by the length of the structure prototype vector to perform inhibitory adjustment, a disturbance penalty term is constructed, and a structure selective mapping weight is obtained, and the specific mathematical model is: ; In the formula, is a structure selective mapping weight of the i-th slice corresponding to the j-th structure prototype vector at scale z, is a structure selective gating tensor, is a tower crane tower feature embedding, is each element in a structure disturbance tensor representation matrix . is the non-negative truncation function symbol, In order to control the power exponent of the periodic response amplitude intensity, the tower crane tower reconstruction result is obtained by weighting and summing all structural prototype vectors with the structure selective mapping weight. The specific mathematical expression is: ; Where, is the number of structure prototype vectors, is the structure prototype vector index, The tower crane tower reconstruction results at different scales are fused to form a unified tower crane tower time slice feature. , and then reconstructed into the tower crane tower body sequence through the decoder , calculate the selected slice mapping score, and measure the deviation between the tower crane tower feature embedding and the tower crane tower reconstruction result through Euclidean distance. The specific mathematical model is: ; Where, Score the selected slice map, is the normalization coefficient, is the number of different scales involved in the calculation, is the scale index.

[0019] Preferably, by constructing a time-variable phase evolution trajectory tensor, the phase change trajectory of the tower body's multivariable state within the periodic window is extracted, and a phase evolution induced graph is further generated to capture the synchronization structure between variables in different time segments. The current periodic structure is structurally mapped by constructing a set of prototype graphs, and the matching deviation between the current periodic graph and all prototypes is used as the periodic anomaly score. The periodic anomaly identification is completed in combination with the threshold judgment mechanism. This design integrates the ideas of graph structure learning and phase evolution modeling, breaking through the traditional periodic detection method's dependence on time domain trends and the limitations of independent variable modeling, and realizing the structured, high-dimensional expression and accurate discrimination of tower body periodic anomalies.

[0020] Preferably, in the operational monitoring of the tower crane tower structure, its structural state evolves continuously over time, and the health characteristics at different times often have significant dynamic changes and trend deviations. The traditional static weight mechanism cannot adapt to the non-stationarity of time series characteristics, and easily ignores the impact of historical information on the current state, resulting in a delayed response to structural changes, especially in the early stages of abnormal state evolution, when the amplitude of the change is weak and difficult to accurately capture. Therefore, it is urgent to build a dynamic and adaptive weight update mechanism that can integrate historical weights with current feature information, autonomously adjust the rate and amplitude of weight changes, and achieve sensitive tracking of health state fluctuations.

[0021] Preferably, in S31, the input is also the tower sequence of the reduction tower crane , and the specific mathematical model is: ; In the formula, is the tower sequence of the reduction tower crane of the t time step and the c variable, is a phase angle operation on complex numbers, is a Hilbert transform operation, is the corresponding phase orbit tensor of the t time step and the c variable, the phase evolution induction map is constructed, the sliding window operation is performed on the phase orbit tensor, the cosine value of the phase trajectory difference of the phase orbit tensor in the time period is calculated, the results in the entire window length are averaged, and the phase evolution induction map reflecting the synchronization behavior between the phase orbit tensors in the time window is generated, and the specific mathematical model is: ; In the formula, is the phase evolution induction map of the i and j phase orbit tensors in the l sliding window, is the length of the sliding window, , is the phase orbit tensor of the i and j variables in the l sliding window, is a sliding window, is the time step index in the sliding window.

[0022] Preferably, by constructing an adaptive time sequence weight updating mechanism, a difference adjustment term between historical weights and current features is introduced, a nonlinear mapping function and multiple trainable parameters are combined to control the weight change trend, so as to realize flexible adjustment of the weight updating rate and amplitude, and by adjusting the parameters, the current feature response is suppressed or enhanced, so that the model has stronger adaptive ability when facing different health state fluctuations. This design breaks through the limitation of traditional static or linear weight distribution strategy on the flexibility of time series modeling, significantly enhances the perception ability of the model to the dynamic evolution trend of the tower health features, and provides time sequence continuity guarantee for subsequent health value prediction.

[0023] When modeling the health status of a tower crane, relying solely on feature information at the current moment often fails to fully reflect the underlying evolution of the structural state. Historical information, however, plays a crucial role in identifying trends in state changes. Traditional methods often directly concatenate or average historical and current features, failing to consider the relative weights and nonlinear interactions between features at different times. This can easily lead to information redundancy or feature degradation, impacting the accuracy of subsequent predictions. Therefore, it is necessary to develop a mechanism that can rationally integrate historical and current features to achieve multi-moment joint modeling of the structural state, providing more stable and complete feature input for the final health prediction.

[0024] Preferably, in said S32, a prototype graph set consisting of multiple graph structures is defined , the Frobenius distance is used to calculate the structural deviation metric between each phase evolution induced graph and each prototype graph. The specific mathematical model is: ; Where, is the phase evolution induced graph in the lth sliding window The structural deviation measure between the k-th prototype graph, is the phase evolution induced graph of the i-th and j-th phase orbit tensors in the l-th sliding window, is the kth prototype graph in the prototype graph set, is the Frobenius norm, integrating the original structural state and the current induced graph information to design the gating factor Control the update amplitude and fusion method. The specific mathematical model is: ; Where, 、 is the trainable weight matrix, is the activation function, which dynamically adjusts the structural expression of the prototype graph through the gating mechanism. The specific mathematical model is: ; Where, Controlled by the gating factor.

[0025] Preferably, by fusing the historical moment features and the current moment feature sequence on the basis of adaptive weight adjustment, the multi-moment feature information is combined and modeled in a dynamic weight distribution manner, a time sequence enhancement mechanism is introduced in the structure to maintain the time sequence dependency of the features, and the fused features are expressed and enhanced through a nonlinear activation function, thereby further constructing a unified health feature vector. This design not only solves the problems of feature redundancy and information conflict in multiple moments, but also effectively integrates the structural health evolution trend and improves the perception ability of the input features to the future state change, thereby providing a high-quality feature basis for the final monitoring value output.

[0026] Preferably, the change process of the tower crane tower health state usually has nonlinear and multi-factor coupling characteristics, and the traditional linear modeling method is difficult to depict the complex mapping relationship between the input features and the health monitoring value. Especially in the face of sudden disturbance or structural anomaly, the model is not sensitive to the response of key features, resulting in large deviation of the monitoring result. In addition, the feature distribution exhibited by the tower structure in different operation stages is dynamic, and a structure with good nonlinear fitting capability is needed to improve the prediction performance. Therefore, a network structure with nonlinear expression capability and learning capability needs to be introduced to effectively map the fused features and accurately output the health monitoring value.

[0027] Preferably, in the S33, the phase response deviation score is calculated, the corresponding phase evolution induced graph is extracted for each sliding window, and the Frobenius distance is measured based on the distance relationship between the induced graph and a plurality of updated prototype graphs. The specific mathematical model is: ; In the formula, is the phase response deviation score, is the phase evolution induced graph of the i-th and j-th phase orbit tensor in the l-th sliding window, is the k-th prototype graph in the prototype graph set, is the Frobenius norm, is the comparison of all k=1, 2,..., k prototype graphs, and the closest prototype graph distance is selected as the phase response deviation score.

[0028] Preferably, the health feature vector after fusion is transformed in the hidden space by adopting a feature mapping network constructed by a full connection layer combined with a nonlinear activation function, and a health monitoring value is output through a second layer linear transformation, each parameter in the model is automatically learned through training, has strong expression flexibility and fitting capacity, can effectively model the nonlinear relationship between the features and the output values, this design breaks through the limitations of traditional linear prediction models in high-dimensional feature modeling, at the same time, the prediction path is constructed combining the previous time sequence structure and dynamic feature information, which significantly improves the prediction accuracy of the tower structure health state and the time sequence response ability of the model.

[0029] Preferably, in the S4, an abnormality judgment module is constructed, the phase response deviation score and the selected slice mapping score are input , a multi-source fusion abnormality score is obtained through weighted fusion , a threshold is set , and the tower crane tower body damage abnormality judgment is completed.

[0030] In summary, due to the adoption of the technical solution, the present application has the following beneficial effects: the present application proposes a tower crane tower body damage abnormality detection method, constructs a CTS-DAM abnormality detection model, and applies it to the tower crane tower body damage abnormality detection scene, which as a whole includes a selected slice mapping module, a phase period abnormality modeling module and an abnormality judgment module, the selected slice mapping module is used for local disturbance analysis of data, the phase response driving module is used for processing periodic abnormality information, the abnormality judgment module is used for integrating two scores to obtain a final abnormality score and complete abnormality judgment, and the modules cooperate with each other to realize abnormality detection of the tower crane tower body damage. BRIEF DESCRIPTION OF DRAWINGS

[0031] Figure 1 It is a tower crane tower body damage abnormality detection method flowchart.

[0032] Figure 2 It is a selected slice mapping module diagram.

[0033] Figure 3 It is a phase period abnormality modeling module diagram.

[0034] Figure 4 It is a tower crane tower body damage abnormality detection effect diagram.

[0035] Figure 5 It is a CTS-DAM abnormality detection model error distribution diagram. DETAILED DESCRIPTION

[0036] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application; based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0037] Please refer to Figures 1-5 , the present application provides a technical solution: a tower crane tower damage anomaly detection method, by constructing the selected slice mapping module for local disturbance analysis of data, phase response driving module for processing periodic abnormal information, abnormal judgment module for integrating two scores, getting the final abnormal score, the specific steps are as shown in Figure 1 .

[0038] S1, collect the relevant data of the tower crane tower, construct the data set, and divide the training set, the validation set and the test set after preprocessing.

[0039] Further, the relevant data of the tower crane tower is collected, including tower vibration sensor data, tower stress sensor data, tower temperature sensor data, tower displacement sensor data and tower ultrasonic wave detection data, a multidimensional data set is constructed, and the mean filling method is used for data preprocessing, and the specific mathematical model is: ; In the formula, is the total number of tower crane tower data, is the average value, is the observation value, and the training set, the validation set and the test set are divided according to the ratio of 7:1:2.

[0040] S21, input the tower crane tower data for multi-scale division, introduce the feature coding method of self-attention mechanism, and calculate the tower crane tower feature embedding.

[0041] Further, the selected slice mapping module is constructed, as shown in Figure 2 , the input tower crane tower data , the input tower crane tower data is processed by setting fixed scale parameters under each scale, and the specific mathematical model is: ; In the formula, is the scale parameter, indicating the slice scale, which is set to [12, 24, 48], is the number of the i-th slice, is the end time step number of the slice, tower data of the tower crane at the t-th time step, tower slice vector of the tower crane at the i-th slice after averaging, for size z, For the mean operation, the feature encoding method introducing self-attention mechanism is used to calculate the tower feature embedding of the tower crane The tower slice vector of the tower crane is mapped to query , key and value , , , where , ; In the formula, is the activation function, is the feature embedding dimension, is the transpose of the key, and the feature embedding of each scale is represented as: ; In the formula, is the similarity score of the i-th query to the j-th key, is the j-th value, is the number of tower crane slice at scale z, is the slice index that is noticed, indicating the position of all keys and values, and the specific implementation code is: batch_size = 8 time_length = 240 feature_dim = 16 input_data = torch.randn(batch_size, time_length, feature_dim) # Define multi-scale division scales = [12, 24, 48] # Different scale window lengths stride_ratio = 0.5 # Sliding window step length ratio def multi_scale_patch(input_seq, scales, stride_ratio): batch_size, time_len, feat_dim = input_seq.shape all_patches = [] for win_size in scales: stride = int(win_size * stride_ratio) num_patches = (time_len - win_size) / / stride + 1 scale_patches = [] for i in range(num_patches): start = i * stride end = start + win_size patch = input_seq[:, start:end, :] # [B, win, F] patch_mean = patch.mean(dim=1) # [B, F] scale_patches.append(patch_mean.unsqueeze(1)) # [B, 1, F] scale_patches = torch.cat(scale_patches, dim=1) # [B, N, F] all_patches.append(scale_patches) return all_patches # List of [B, N_z, F] # 执行切片生成 multi_scale_inputs = multi_scale_patch(input_data, scales, stride_ratio) # 定义自注意力嵌入模块 class MultiHeadAttentionEncoder(nn.Module): def __init__(self, d_model, n_heads, dropout=0.1): super().__init__() self.mha = nn.MultiheadAttention(embed_dim=d_model, num_heads=n_heads, dropout=dropout, batch_first=True) self.norm = nn.LayerNorm(d_model) self.ffn = nn.Sequential( nn.Linear(d_model, d_model * 4), nn.ReLU(), nn.Linear(d_model * 4, d_model) ) self.dropout = nn.Dropout(dropout) def forward(self, x): # Self-Attention attn_output, _ = self.mha(x, x, x) x = self.norm(x + self.dropout(attn_output)) # Feedforward network ffn_output = self.ffn(x) x = self.norm(x + self.dropout(ffn_output)) return x # Instantiate the embedder encoder = MultiHeadAttentionEncoder(d_model=feature_dim, n_heads=4) # Embed structure fragments at all scales scale_embeddings = [] for scale_input in multi_scale_inputs: # Each one is [B, N_z, F] encoded = encoder(scale_input) # Output dimension [B, N_z, F] scale_embeddings.append(encoded) # Merge all scale output features final_embedding = torch.cat(scale_embeddings, dim=1) # [B, sum(N_z),F]

[0042] S22, input the tower crane tower feature embedding to calculate the structure disturbance factor, and construct a structure disturbance tensor representation matrix to obtain a structure selective gate tensor.

[0043] Further, the input tower crane tower feature embedding proposes a disturbance response strategy to reduce all tower crane tower feature embeddings at the current scale to obtain a reference center representation at scale z The specific mathematical expression is: ; In the formula, is the number of tower crane tower slices at scale z, is the index of the slice, and the structural difference of each scale tower crane tower feature embedding relative to the reference center representation is calculated to form a measurable structural disturbance factor. The specific mathematical expression is: ; In the formula, is the structural disturbance factor of the i-th tower crane tower feature embedding at scale z, is the Euclidean distance square, and then based on the response offset relationship between different tower crane tower feature embeddings in the embedding space, a structure disturbance tensor representation matrix is constructed. The specific mathematical expression is: ; In the formula, is the structure disturbance tensor representation matrix, is the softmax normalization of the matching distribution of each slice at each scale z, is the j-th structure prototype vector at scale z, representing a predefined structure template, is a nonlinear mapping function, is the number of structure prototype vectors, set to 16, is the index of the structure disturbance factor, is the structure prototype vector index, by jointly considering the tower crane tower feature embedding of the current slice, the structure prototype vector and the structure disturbance factor, and combining the trainable linear discriminant parameter, a logical judgment is made on whether to activate each structure prototype path to obtain a structure selective gate tensor. The specific mathematical expression is: {G}^{z}_{ij}=\sigma ({W}_{g}\cdot \left [ {{q}^{z}_{i}||{b}^{z}_{j}||{\Delta}^{z}_{i}} \right ]+{b}_{g}) ; where, is an activation function, is a gating weight matrix, is a gating bias term, is a structure-selective gating tensor, and the code is implemented as: B, N, F = 8, 60, 32 # B batch size, N number of segments, F embedding dimension H = 16 # Number of prototypes input_embedding = torch.randn(B, N, F) # Define the structure prototype vector [H, F] struct_prototypes = nn.Parameter(torch.randn(H, F)) # Define the perturbation perception module class StructuralPerturbationMapping(nn.Module): def __init__(self, embed_dim, num_prototypes): super().__init__() self.embed_dim = embed_dim self.num_prototypes = num_prototypes # Structure perturbation factor, trainable [H, F] self.perturbation = nn.Parameter(torch.randn(num_prototypes,embed_dim)) # Gating discriminant network self.gate_fc = nn.Sequential( nn.Linear(embed_dim * 3, embed_dim), nn.ReLU(), nn.Linear(embed_dim, 1) ) # Gating activation function is Sigmoid self.gate_activation = nn.Sigmoid() def forward(self, x, prototypes): B, N, F = x.shape H = self.num_prototypes # Expand dimensions for computation [B, N, 1, F]&[1, 1, H, F] x_expand = x.unsqueeze(2).expand(B, N, H, F) # [B, N, H, F] p_expand = prototypes.unsqueeze(0).unsqueeze(0).expand(B, N, H,F) # [B, N, H, F] s_expand = self.perturbation.unsqueeze(0).unsqueeze(0).expand(B,N, H, F) # Construct perturbation response tensor (element-wise difference) delta = x_expand - p_expand # [B, N, H, F] perturb_tensor = delta * s_expand # Perturbation weighted # Build gating representation: input [x, prototype, perturbation] concat_feat = torch.cat([x_expand, p_expand, perturb_tensor], dim=-1) # [B, N, H, 3F] gate_score = self.gate_fc(concat_feat).squeeze(-1) # [B,N, H] gate_weight = self.gate_activation(gate_score) # ∈(0,1), can be used as structural selective gating tensor # Build non-normalized structural mapping weights (considering perturbation response intensity) sim_score = F.cosine_similarity(x_expand, p_expand, dim=-1)# [B, N, H] penalty = torch.norm(perturb_tensor, dim=-1) # [B, N,H] mapping_weight = sim_score - penalty * gate_weight # non-normalized weight return mapping_weight, gate_weight, perturb_tensor # Instantiate the structure mapping module structure_mapper = StructuralPerturbationMapping(embed_dim=F, num_prototypes=H) # Execute structural perturbation mapping mapping_weight, gate_weight, perturb_tensor = structure_mapper(input_embedding, struct_prototypes)

[0044] S23. Based on the relationship between the structural perturbation factor and the structural prototype vector, multiple coupling-driven global scaling factors are designed.

[0045] Furthermore, multiple coupled-driven global scaling factors are designed. Based on the relationship between the structural perturbation factor and the structural prototype vector, the nonlinear enhancement term and the periodic modulation factor are integrated to obtain the scaling factor corresponding to each slice through coupled modeling. The specific mathematical model is: ; Where, is the multi-coupling driven global scaling factor for the i-th slice at scale z, is the amplification factor of the polynomial modulation term, initially set to 2.5, is the structural perturbation factor of the i-th tower crane tower feature embedding at scale z, is the polynomial function of the perturbation factor of the i-th slice structure, To control the scaling factor of the periodic term frequency, the initial setting is 1.8. is the frequency control index of the i-th slice, To control the phase shift coefficient, the initial setting is 0.5. For the scale z, the structure prototype vector The average similarity with the current query, For the power index of the control period response amplitude, initially set to 1.2, and the specific implementation code is: B, N = 8, 60 t_index = torch.linspace(0, 2 * math.pi, N).unsqueeze(0).repeat(B, 1) # Time phase input # Hyperparameters lambda_poly = 2.5 # Polynomial modulation amplification factor omega_scale = 1.8 # Scaling factor of the control period term frequency phi_shift = math.pi / 4 # Phase shift gamma_power = 1.2 # Power index of the period response amplitude # Define the periodic modulation module class PeriodicScalingModule(nn.Module): def __init__(self, lambda_poly, omega_scale, phi_shift, gamma_power): super().__init__() self.lambda_poly = lambda_poly self.omega_scale = omega_scale self.phi_shift = phi_shift self.gamma_power = gamma_power def forward(self, t): # Trigonometric function modulation term trig_term = torch.sin(self.omega_scale * t + self.phi_shift) #[B, N] # Polynomial modulation term (power index + amplification) poly_term = (1 + self.lambda_poly * t**2) # [B, N] Coupling modulation term: polynomial modulation superimposed periodic response scale_factor = poly_term * torch.abs(trig_term) ** self.gamma_power # [B, N] return scale_factor

[0046] S24, introduce S21, S22 output and multi-coupling driven global scaling factor, calculate structure selective mapping weight, reconstruct the tower crane tower sequence, and obtain the final slice mapping score.

[0047] Further, a structure selective mapping weight generation strategy is proposed to fuse multi-factor regulation. The structure selective mapping weight is constructed, the matching relationship between each slice and the structure prototype vector is scaled and adjusted based on the multi-coupling driven global scaling factor, the matching strength after scaling is weighted slice-specifically combined with the structure selective gating tensor, and then the matching similarity between the slice and the structure prototype vector is introduced as the basic structure similarity expression. The product of each element in the structure disturbance tensor representation matrix and the length of the structure prototype vector is suppressed and adjusted to construct the disturbance penalty term, and the structure selective mapping weight is obtained. The specific mathematical model is: ; In the formula, is the structure selective mapping weight of the i-th slice corresponding to the j-th structure prototype vector at the scale z, is the structure selective gating tensor, is the tower crane tower feature embedding, is each element in the structure disturbance tensor representation matrix , is a non-negative truncation function symbol, is a power index that controls the amplitude of the periodic response. By weighting and summing all structure prototype vectors with structure selective mapping weights, the tower crane tower reconstruction result is obtained. The specific mathematical expression is: ; In the formula, is the number of structure prototype vectors, which is set to 16, is the structure prototype vector index, is the tower crane tower reconstruction result. The tower crane tower reconstruction results at different scales are fused to form a unified tower crane tower time slice feature , and then reconstructed into the restored tower crane tower sequence , calculate the selected slice mapping score, and measure the deviation between the tower crane tower feature embedding and the tower crane tower reconstruction result by Euclidean distance. The specific mathematical model is: ; In the formula, is the selected slice mapping score, is the normalization coefficient, set to 1, is the number of different scales participating in the calculation, set to 32, is the scale index, and the specific implementation code is: B, N = 8, 60 H, F = 16, 32 # Structure deviation score [B, N] structure_deviation_score = torch.rand(B, N) # Periodic anomaly response tensor [B, N] temporal_anomaly_response = torch.rand(B, N) # Time index sequence, construct periodic modulation term [B, N] t_index = torch.linspace(0, 2 * math.pi, N).unsqueeze(0).repeat(B, 1) class CoupledPeriodicScaler(nn.Module): def __init__(self, lambda_poly=2.5, omega=2.2, phi=math.pi / 6, gamma=1.5): super().__init__() self.lambda_poly = lambda_poly self.omega = omega self.phi = phi self.gamma = gamma def forward(self, t): # t: [B, N] trig_term = torch.sin(self.omega * t + self.phi) # Trigonometric periodic modulation poly_term = 1 + self.lambda_poly * t ** 2 # polynomial modulation coupled = poly_term * torch.abs(trig_term) ** self.gamma # coupling enhancement return coupled # [B, N] class PeriodicAnomalyDiscriminator(nn.Module): def __init__(self, embed_dim=32, use_transformer=False): super().__init__() self.scaler = CoupledPeriodicScaler() # Anomaly score fusion network if use_transformer: encoder_layer = nn.TransformerEncoderLayer(d_model=embed_dim,nhead=4, batch_first=True) self.encoder = nn.TransformerEncoder(encoder_layer, num_layers=2) self.project = nn.Linear(embed_dim, 1) else: self.encoder = None self.project = nn.Sequential( nn.Linear(2, 32), nn.ReLU(), nn.Linear(32, 1) ) def forward(self, struct_dev, phase_resp, t_idx): # Get the period scaling factor [B, N] scale = self.scaler(t_idx) # Period deviation term = Period response * Scaling factor phase_scaled = phase_resp * scale # [B, N] # Fusion structure and period deviation [B, N, 2] fusion = torch.stack([struct_dev, phase_scaled], dim=-1) # Encoder modeling if self.encoder is not None: encoded = self.encoder(fusion) anomaly_logits = self.project(encoded).squeeze(-1) # [B, N] else: logits = self.project(fusion) # [B, N, 1] anomaly_logits = logits.squeeze(-1) # Final period anomaly score anomaly_score = torch.mean(anomaly_logits, dim=1) # [B] return anomaly_score, anomaly_logits, scale

[0048] S31、Input the restored tower crane tower sequence, project to obtain the phase orbit tensor, perform sliding window operation and construct the phase evolution induction map.

[0049] Further, a phase period anomaly modeling module is constructed, as shown in Figure 3 , input the restored tower crane tower sequence , project each restored tower crane tower sequence to the Hilbert phase domain, and the specific mathematical model is: ; In the formula, is the restored tower crane tower sequence of the t-th time step and the c-th variable, is a phase angle operation taking complex numbers, is a Hilbert transform operation, For the phase orbit tensor corresponding to the cth variable at the tth time step, a phase evolution induction map is constructed. A sliding window operation is performed on the phase orbit tensor. By calculating the cosine value of the phase trajectory difference of the phase orbit tensor within the time period, the results within the entire window length are averaged to generate a phase evolution induction map reflecting the synchronization behavior between the phase orbit tensors under the time window. The specific mathematical model is: ; Where, is the phase evolution induced graph of the i-th and j-th phase orbit tensors in the l-th sliding window, is the length of the sliding window, set to 120, 、 is the phase orbit tensor of the i-th and j-th variables in the l-th sliding window, is a sliding window, is the time step index within the sliding window. The specific implementation code is: B, C, T = 8, 5, 120 X_enc = torch.randn(B, C, T) # Features after encoder output [B, C, T] def extract_phase_trajectory_hilbert(x_tensor): phase_tensor = [] for b in range(x_tensor.shape[0]): batch_phase = [] for c in range(x_tensor.shape[1]): signal_np = x_tensor[b, c].detach().cpu().numpy() analytic = hilbert(signal_np) phase = np.angle(analytic) # [-π, π] batch_phase.append(phase) phase_tensor.append(batch_phase) phase_tensor = np.array(phase_tensor) # [B, C, T] return torch.tensor(phase_tensor, dtype=torch.float32) def extract_phase_trajectory_stft(x_tensor, window_size=32, hop_length=8): B, C, T = x_tensor.shape phase_tensor = [] for b in range(B): batch_phase = [] for c in range(C): signal = x_tensor[b, c].detach().cpu().numpy() Zxx = scipy.signal.stft(signal, nperseg=window_size, noverlap=window_size - hop_length) phase = np.angle(Zxx) # [Freq, Time] # 取主频分量的相位 phase_avg = np.mean(phase[:3, :], axis=0) # [T'] # 插值回原始长度 T phase_interp = np.interp(np.linspace(0, len(phase_avg) - 1, T),np.arange(len(phase_avg)), phase_avg) batch_phase.append(phase_interp) phase_tensor.append(batch_phase) phase_tensor = np.array(phase_tensor) return torch.tensor(phase_tensor, dtype=torch.float32) def compute_phase_tensor(X_enc, method='hilbert'): if method == 'hilbert': return extract_phase_trajectory_hilbert(X_enc) elif method == 'stft': return extract_phase_trajectory_stft(X_enc) else: raise ValueError("Unsupported phase method")

[0050] S32. Define a set of prototype graphs, calculate the structural deviation measure between the phase evolution induced graph and the prototype graph, and design a gating factor to dynamically adjust the structural expression of the prototype graph through the gating mechanism.

[0051] Furthermore, a prototype graph set consisting of multiple graph structures is defined , the Frobenius distance is used to calculate the structural deviation metric between each phase evolution induced graph and each prototype graph. The specific mathematical model is: ; Where, is the phase evolution induced graph in the lth sliding window The structural deviation measure between the k-th prototype graph, is the phase evolution induced graph of the i-th and j-th phase orbit tensors in the l-th sliding window, is the kth prototype graph in the prototype graph set, is the Frobenius norm, integrating the original structural state and the current induced graph information to design the gating factor Control the update amplitude and fusion method. The specific mathematical model is: ; Where, 、 is the trainable weight matrix, is the activation function, which dynamically adjusts the structural expression of the prototype graph through the gating mechanism. The specific mathematical model is: ; Where, For gate factor control, the specific implementation code is: def compute_phase_diff_cosine(phase_seq): B, C, W = phase_seq.shape G = torch.zeros(B, C, C) for b in range(B): for i in range(C): for j in range(C): diff = phase_seq[b, i] - phase_seq[b, j] # Phase difference [W] cos_sim = torch.mean(torch.cos(diff)) # Phase synchronization G[b, i, j] = cos_sim return G # [B, C, C] def build_induced_graphs(phase_tensor, window_size=15, hop=1): B, C, T = phase_tensor.shape G_list = [] # Sliding window traversal for t in range(0, T - window_size + 1, hop): window = phase_tensor[:, :, t:t + window_size] # [B, C, W] G = compute_phase_diff_cosine(window) # [B, C, C] G_list.append(G) G_stack = torch.stack(G_list, dim=1) # [B, T', C, C] # Fill to the original time length T pad_len = T - G_stack.shape[1] if pad_len>0: pad_tensor = G_stack[:, :1, :, :].repeat(1, pad_len, 1, 1) G_full = torch.cat([pad_tensor, G_stack], dim=1) # [B, T, C, C] else: G_full = G_stack return G_full

[0052] S33. Obtain a phase response deviation score by calculating the distance relationship between the induced image and multiple updated prototype images.

[0053] Furthermore, the phase response deviation score is calculated, and the corresponding phase evolution induced map is extracted for each sliding window. The Frobenius distance is measured based on the distance relationship between the induced map and multiple updated prototype maps. The specific mathematical model is: ; Where, is the phase response deviation score, is the phase evolution induced graph of the i-th and j-th phase orbit tensors in the l-th sliding window, is the kth prototype graph in the prototype graph set, is the Frobenius norm, To compare all prototype images with k=1,2,...,k, the closest prototype image distance is selected as the phase response deviation score. The specific implementation code is: class StructuralPrototypeMatcher(nn.Module): def __init__(self, num_prototypes=6, num_vars=5): super().__init__() self.K = num_prototypes self.C = num_vars # Prototype parameters self.prototypes = nn.Parameter(torch.randn(self.K, self.C,self.C)) self.symmetrize = True # Graph Attention Layer self.attention_layer = nn.MultiheadAttention(embed_dim=self.C *self.C, num_heads=2) def forward(self, A_seq): B, T, C, _ = A_seq.shape K = self.K # Flatten the image to a vector [B, T, C*C] A_flat = A_seq.reshape(B, T, -1) # [B, T, C*C] G_flat = self.prototypes.reshape(K, -1) # [K, C*C] # Optional: Graph Attention Matching A_attn_input = A_flat.permute(1, 0, 2) # [T, B, C*C] G_attn_input = G_flat.unsqueeze(1).repeat(1, B, 1) # [K, B, C*C] G_attn_input = G_attn_input.permute(2, 1, 0) # [C*C, B, K] # Attention Enhanced Matching _, _ = self.attention_layer(A_attn_input, G_attn_input, G_attn_input) # Ignore output to enhance learnability # Calculate Frobenius distance as deviation score A_exp = A_seq.unsqueeze(2) # [B, T, 1, C, C] G_exp = self.prototypes.unsqueeze(0).unsqueeze(0) # [1, 1, K, C,C] diff = A_exp - G_exp # [B, T, K, C, C] dist = torch.norm(diff, p='fro', dim=(3, 4)) # [B, T, K] scores, _ = torch.min(dist, dim=2) # [B, T] return scores

[0054] S4. Construct an abnormality judgment module, perform weighted fusion of the phase response deviation score and the selected slice mapping score, and set a threshold to judge abnormal damage of the tower crane tower body.

[0055] Furthermore, an abnormality judgment module is constructed, and the phase response deviation score is input Slice mapping score with knot selection , weighted fusion to obtain multi-source fusion anomaly score , by setting the threshold , set to 25, to complete the abnormal judgment of tower crane tower damage.

[0056] Furthermore, the CTS-DAM anomaly detection model is written in Python. The experiment runs on the Windows operating system. Pytorch is selected as the framework in the CUDA11.27 environment. Training is performed on a GeForce RTX 3090. The optimizer is selected, the initial learning rate is set to 0.001, the training batch is set to 64, the training cycle is set to 100, and the dataset is 60 days of crane tower-related data, which is input into the CTS-DAM anomaly detection model after preprocessing.

[0057] Furthermore, the error distribution diagram of the CTS-DAM anomaly detection model and the effect diagram of tower crane tower damage anomaly detection are shown in the figure below. Figure 4 、 Figure 5 As shown, in Figure 4 The abnormal characteristic values ​​detected in the data fluctuate significantly over time, especially in the 7th and 11th months, with high peaks respectively, and significantly exceeding the warning threshold and abnormal threshold, indicating that the system is more sensitive to structural damage identification at these two time points; at the same time, the abnormal characteristic values ​​in other time periods remain at a low level, reflecting the model's robust response ability to normal conditions. Figure 5 The majority of the model errors are concentrated near zero, forming a relatively symmetrical distribution. This indicates that the detection model has a small overall prediction error, limited deviation, and good stability and robustness. Furthermore, by setting a zero-error boundary, the error drift trend can be intuitively observed, further assessing the degree of response deviation of the model under different states. The combined results of the two figures demonstrate that the constructed structural damage anomaly detection model has high accuracy and effective anomaly identification capabilities.

Claims

1. A tower crane tower damage anomaly detection method, characterized in that: The following steps are involved: S1. Collect tower crane tower body related data, construct a dataset, and divide it into training set, validation set and test set after preprocessing; S2. Construct a knot selection slice mapping module for local perturbation analysis. The specific steps are as follows: S21. The input tower crane body data is divided into multiple scales, a feature encoding method based on the self-attention mechanism is introduced, and the tower crane body feature embedding is calculated; S22, input tower crane tower body feature embedding to calculate the structural perturbation factor, and construct the structural perturbation tensor representation matrix to obtain the structural selective gating tensor; S23. Based on the relationship between the structural perturbation factor and the structural prototype vector, design multiple coupled driving global scaling factors; S24, introduce the outputs of S21 and S22 and multiple coupling-driven global scaling factors, calculate the structure selective mapping weight, reconstruct and generate the restored tower crane tower body sequence, and obtain the knot selection slice mapping score; S3. Build a phase response driving module to process periodic abnormal information. The specific steps are as follows: S31. Input the restored tower crane tower body sequence, project it to obtain the phase orbit tensor, perform a sliding window operation, and construct a phase evolution induced graph; S32. Define a set of prototype graphs, calculate the structural deviation metric between the phase evolution induced graph and the prototype graph, and design a gating factor to dynamically adjust the structural expression of the prototype graph through the gating mechanism; S33, obtaining a phase response deviation score by calculating a distance relationship between the induced image and multiple updated prototype images; S4. Construct an abnormality judgment module, perform weighted fusion of the phase response deviation score and the selected slice mapping score, and set a threshold to judge abnormal damage of the tower crane tower body.

2. A tower crane tower body damage anomaly detection method according to claim 1, characterized in that: In said S21, the tower crane tower body data is input , by setting a fixed scale parameter , the input tower crane tower data is divided into non-overlapping mean values ​​at each scale. The specific mathematical model is: ； Where, is the scale parameter, indicating the slice scale, is the number of the i-th slice, is the ending time step number of the slice, is the tower crane tower data at the tth time step, is the tower crane tower slice vector after averaging the i-th slice under size z, In order to obtain the mean value, the feature encoding method of the self-attention mechanism is introduced to calculate the tower crane tower feature embedding , the tower crane tower slice vector is mapped to the query through three independent linear transformations ,key Sum ,in 、 、 is a learnable weight matrix, and then self-attention calculates the similarity score by matching the query with the key. The specific mathematical model is: ； Where, is the activation function, is the feature embedding dimension, With t being the transpose of the key, the feature embedding at each scale is expressed as: ； Where, is the similarity score of the i-th query to the j-th key, is the jth value, is the number of tower crane tower slices at scale z, The slice index being watched indicates the location of all keys and values.

3. A tower crane tower body damage anomaly detection method according to claim 2, characterized in that: In S22, the tower crane body feature is input and embedded. , a disturbance response strategy is proposed to embed all tower crane tower features at the current scale and perform reduction processing to obtain the reference center representation at scale z. , the specific mathematical expression is: ； Where, is the number of tower crane tower slices at scale z, The index of the slice is used to calculate the structural difference of the tower crane tower feature embedding at each scale relative to the reference center to form a measurable structural perturbation factor. The specific mathematical expression is: ； Where, is the structural perturbation factor of the i-th tower crane tower feature embedding at scale z, is the square of the Euclidean distance. Then, based on the response offset relationship between the tower crane feature embeddings in the embedding space, the structural disturbance tensor representation matrix is ​​constructed. The specific mathematical expression is: ； Where, is the structural perturbation tensor representation matrix, To perform softmax normalization on the matching distribution of each slice at each scale z, is the j-th structural prototype vector at scale z, representing the predefined structural template, is a nonlinear mapping function, is the number of structure prototype vectors, is the index of the structural perturbation factor, is the structural prototype vector index. By jointly considering the tower crane tower feature embedding of the current slice, the structural prototype vector, and the structural perturbation factor, and combining the trainable linear discriminant parameter, a logical judgment is made on whether to activate each structural prototype path to obtain the structural selective gating tensor. The specific mathematical expression is: ； Where, is the activation function, is the gating weight matrix, is the gate bias term, is the structure-selective gating tensor.

4. A tower crane tower body damage anomaly detection method according to claim 3, characterized in that: In S23, multiple coupled driving global scaling factors are designed. Based on the relationship between the structural perturbation factor and the structural prototype vector, the nonlinear enhancement term and the periodic modulation factor are integrated to obtain the scaling factor corresponding to each slice through coupled modeling. The specific mathematical model is: ； Where, is the multi-coupling driven global scaling factor for the i-th slice at scale z, is the amplification factor of the polynomial modulation term, is the structural perturbation factor of the i-th tower crane tower feature embedding at scale z, is the polynomial function of the perturbation factor of the i-th slice structure, is the scaling factor for controlling the frequency of the periodic term, is the frequency control index of the i-th slice, To control the phase shift coefficient, is the structural prototype vector at scale z The average similarity with the current query, is the power exponent that controls the amplitude strength of the periodic response.

5. A tower crane tower body damage anomaly detection method according to claim 4, characterized in that: In the S24, a strategy for generating structure selective mapping weights by integrating multi-factor regulation is proposed to construct structure selective mapping weights. The matching relationship between each slice and the structural prototype vector is scaled and adjusted based on multiple coupling-driven global scaling factors. The scaled matching strength is weighted slice-specifically in combination with the structure selective gating tensor. Then, the matching similarity between the slice and the structural prototype vector is introduced as the basic structural similarity expression. The matrix is ​​represented by the structural perturbation tensor. The product of each element in and the modulus of the structure prototype vector is suppressed and regulated to construct a disturbance penalty term and obtain the structure selective mapping weight. The specific mathematical model is: ； Where, is the structure selective mapping weight of the i-th slice corresponding to the j-th structure prototype vector at scale z, is the structure-selective gating tensor, Embed the tower crane tower feature. is the structural perturbation tensor representation matrix For each element in is the non-negative truncation function symbol, In order to control the power exponent of the periodic response amplitude intensity, the tower crane tower reconstruction result is obtained by weighting and summing all structural prototype vectors with the structure selective mapping weight. The specific mathematical expression is: ； Where, is the number of structure prototype vectors, is the structure prototype vector index, The tower crane tower reconstruction results at different scales are fused to form a unified tower crane tower time slice feature. , and then reconstructed into the tower crane tower body sequence through the decoder , calculate the selected slice mapping score, and measure the deviation between the tower crane tower body feature embedding and the tower crane tower body reconstruction result through Euclidean distance. The specific mathematical model is: ； Where, Score the selected slice map, is the normalization coefficient, is the number of different scales involved in the calculation, is the scale index.

6. A tower crane tower body damage anomaly detection method according to claim 5, characterized in that: In said S31, input the sequence of restoring the tower crane tower body , project each restored tower crane tower body sequence into the Hilbert phase domain. The specific mathematical model is: ； Where, is the restored tower crane tower body sequence of the cth variable at the tth time step, To obtain the complex phase angle operation, is the Hilbert transform operation, For the phase orbit tensor corresponding to the cth variable at the tth time step, a phase evolution induction map is constructed. A sliding window operation is performed on the phase orbit tensor. By calculating the cosine value of the phase trajectory difference of the phase orbit tensor within the time period, the results within the entire window length are averaged to generate a phase evolution induction map reflecting the synchronization behavior between the phase orbit tensors under the time window. The specific mathematical model is: ； Where, is the phase evolution induced graph of the i-th and j-th phase orbit tensors in the l-th sliding window, is the length of the sliding window, 、 is the phase orbit tensor of the i-th and j-th variables in the l-th sliding window, is a sliding window, is the time step index within the sliding window.

7. A tower crane tower body damage anomaly detection method according to claim 6, characterized in that: In the above S32, a prototype graph set consisting of multiple graph structures is defined. , the Frobenius distance is used to calculate the structural deviation metric between each phase evolution induced graph and each prototype graph. The specific mathematical model is: ； Where, is the phase evolution induced graph in the lth sliding window The structural deviation measure between the k-th prototype graph, is the phase evolution induced graph of the i-th and j-th phase orbit tensors in the l-th sliding window, is the kth prototype graph in the prototype graph set, is the Frobenius norm, integrating the original structural state and the current induced graph information to design the gating factor Control the update amplitude and fusion method. The specific mathematical model is: ； Where, 、 is the trainable weight matrix, is the activation function, which dynamically adjusts the structural expression of the prototype graph through the gating mechanism. The specific mathematical model is: ； Where, Controlled by the gating factor.

8. A tower crane tower body damage anomaly detection method according to claim 7, characterized in that: In S33, the phase response deviation score is calculated, and the corresponding phase evolution induced map is extracted for each sliding window. The Frobenius distance is measured based on the distance relationship between the induced map and multiple updated prototype maps. The specific mathematical model is: ； Where, is the phase response deviation score, is the phase evolution induced graph of the i-th and j-th phase orbit tensors in the l-th sliding window, is the kth prototype graph in the prototype graph set, is the Frobenius norm, To compare all prototype images with k=1,2,...,k, the closest prototype image distance is selected as the phase response deviation score.

9. A tower crane tower body damage anomaly detection method according to claim 8, characterized in that: In S4, an abnormality judgment module is constructed and the phase response deviation score is input. Slice mapping score with knot selection , weighted fusion to obtain multi-source fusion anomaly score , by setting the threshold , complete the abnormal judgment of tower crane tower damage.

Images