Digital twin-based fault diagnosis method for logistics mobile equipment

Abstract

Disclosed is a digital twin-based fault diagnosis method for logistics mobile equipment, capable of achieving online monitoring of equipment state and fault diagnosis by means of fusion of digital twin technology and sensor data, thereby allowing for proactive fault prevention and precise maintenance, improving the reliability and operation efficiency of logistics equipment and prolonging the service life thereof. The technical solution of the method comprises: step 1, on the basis of logistics mobile equipment, establishing a corresponding digital twin model; step 2, acquiring real-time data of a battery entity for adaptive model updating; step 3, inputting fault information into a bidirectional GRU model having a multi-head attention mechanism; step 4, performing outlier detection on the basis of an improved fuzzy isolation forest algorithm, comprising introducing a type-2 fuzzy theory into a fuzzy isolation forest algorithm and making improvements; and step 5, using the obtained fault data to implement corrective measures on the basis of risk levels.

Description

Fault diagnosis method for logistics mobile equipment based on digital twinning Field of the invention

[0001] The present invention relates to equipment fault diagnosis, in particular to a fault diagnosis method for logistics mobile equipment based on digital twinning. BACKGROUND

[0002] Mobile equipment in the logistics industry plays a crucial role in ensuring the flow of goods, warehouse management and other links, and is an indispensable key equipment in the modern logistics industry. However, long-term high-load operation and complex working environment make these devices prone to failure, affecting the flow of goods and warehouse efficiency, and increasing maintenance costs. Therefore, developing an efficient and accurate fault diagnosis method is crucial to improve equipment reliability and reduce maintenance costs.

[0003] Currently, the fault diagnosis for logistics mobile equipment is using traditional fault prediction methods, such as statistical-based fault detection and machine learning algorithms. However, these methods often face problems such as complex data processing and low prediction accuracy. Traditional methods usually rely on manual setting of feature extraction rules and model parameters, which can make data processing complex and difficult when dealing with complex equipment working conditions, thus requiring a lot of domain knowledge and professional skills, which increases development and maintenance costs. Due to the complex equipment operating conditions and environmental changes, the pre-set rules and models may not accurately capture potential fault patterns, resulting in unreliable prediction results.

[0004] SUMMARY

[0005] The following presents a simplified summary of one or more aspects in order to provide a basic understanding of such aspects. This summary is not an extensive overview of all contemplated aspects, and is intended to neither identify key or critical elements of all aspects nor delineate the scope of any or all aspects. Its sole purpose is to present some concepts of one or more aspects in a simplified form as a prelude to the more detailed description that is presented later.

[0006] The purpose of the present invention is to solve the above problems, and a fault diagnosis method for logistics mobile equipment based on digital twinning is provided, which realizes online monitoring and fault diagnosis of equipment state through digital twinning technology and sensor data fusion, so as to timely prevent faults and carry out accurate maintenance, improve the reliability, operation efficiency and service life of logistics equipment.

[0007] The technical scheme of the present invention is: the present invention discloses a fault diagnosis method for logistics mobile equipment based on digital twinning, the method comprising:

[0008] Step 1: establishing a corresponding digital twinning model based on logistics mobile equipment;

[0009] Step 2: Collect real-time data of the battery entity for model adaptive update;

[0010] Step 3: Input the fault information into a bidirectional GRU model with a multi-head attention mechanism;

[0011] Step 4: Perform abnormal point detection in combination with an improved fuzzy isolation forest algorithm, including introducing bivalent fuzzy theory and improvement in the fuzzy isolation forest algorithm;

[0012] Step 5: Use the obtained fault data to implement solving measures according to the risk level.

[0013] According to an embodiment of the fault diagnosis method for the logistics mobile equipment based on the digital twin, the model to be established in step 1 is a digital twin model of the battery of the AGV equipment, including the following processing steps:

[0014] Step 1.1: Collect the contour point cloud data of the AGV vehicle body, and pre-process the point cloud data to remove noise and register the point cloud;

[0015] Step 1.2: Perform curve fitting on the pre-processed point cloud data using a NURBS fitting algorithm to obtain a NURBS curve representation of the AGV vehicle body;

[0016] Step 1.3: In the NURBS fitting process, control the positions and values of the control points and weights to affect the shape of the NURBS curve;

[0017] Step 1.4: Form a NURBS surface of the vehicle body surface by connecting the control points of the NURBS curve obtained in step 1.3, thereby realizing 3D model reconstruction of the vehicle body.

[0018] According to an embodiment of the fault diagnosis method for the logistics mobile equipment based on the digital twin, step 2 further includes:

[0019] Step 2.1: Data collection;

[0020] Step 2.2: Perform outlier detection preprocessing on the collected data.

[0021] According to an embodiment of the fault diagnosis method for the logistics mobile equipment based on the digital twin, step 3 further includes:

[0022] Step 3.1: Input the collected AGV battery data in the digital twin model into a bidirectional GRU network;

[0023] Step 3.2: Forward GRU calculation;

[0024] Step 3.3: Reverse GRU calculation;

[0025] Step 3.4: Weighting of the GRU hidden state;

[0026] Step 3.5: Weighted sum of the extracted forward and backward GRU hidden states, and output state information formula;

[0027] Step 3.6: Use the multi-head attention mechanism to calculate the attention weight of the forward GRU hidden state, and then weighted sum with the context vector obtained by the multi-head attention mechanism to obtain the updated forward GRU hidden state.

[0028] According to an embodiment of the fault diagnosis method of the logistics mobile device based on digital twinning, step 4 further comprises:

[0029] Step 4.1: For each feature in the data generated by the GRU network, use an envelope function to represent the membership function;

[0030] Step 4.2: Prepare training set data;

[0031] Step 4.4: Divide the data;

[0032] Step 4.5: Calculate the average path length;

[0033] Step 4.6: Calculate the anomaly score, calculate the average of the normalized path length to get the final anomaly score;

[0034] Step 4.7: Calculate the two-type fuzzy membership, calculate the upper membership and lower membership of each test sample according to the anomaly score and the average path length;

[0035] Step 4.8: Determine the factor set of the evaluation object;

[0036] Step 4.9: Fuzzy comprehensive evaluation result analysis.

[0037] According to an embodiment of the fault diagnosis method of the logistics mobile device based on digital twinning, step 5 further comprises:

[0038] Step 5.1: Divide the hierarchy;

[0039] Step 5.2: For each level and factor in the hierarchy structure, construct a judgment matrix;

[0040] Step 5.3: Calculate the characteristic vector;

[0041] Step 5.4: Calculate the risk level.

[0042] The present application has the following beneficial effects compared with the prior art: the present application compares and analyzes the digital twin model with the actual equipment data, extracts the fault features, and improves the GRU (Gated Recurrent Unit) model for fault diagnosis and prediction. The innovation lies in fully utilizing the advantages of digital twin technology, fusing the actual equipment with the virtual model, and combining the improved GRU network to realize real-time monitoring and fault prediction of the equipment state. Compared with traditional methods, the present application can more accurately predict the fault risk, perform preventive maintenance in advance, significantly reduce the equipment failure rate and maintenance cost, and thus effectively improve the reliability and service life of the equipment. The technology of the present application has important application value and broad market prospect in the efficient operation of the logistics industry. BRIEF DESCRIPTION OF DRAWINGS

[0043] The above features and advantages of the present application can be better understood after reading the detailed description of embodiments of the present application in conjunction with the following drawings. In the drawings, components are not necessarily drawn to scale, and components having similar related properties or features can have the same or similar reference numerals.

[0044] FIG. 1 shows a flowchart of an embodiment of a fault diagnosis method for a logistics mobile equipment based on digital twin of the present application.

[0045] FIG. 2 shows a flowchart of the evolution of the digital twin model in the embodiment shown in FIG. 1.

[0046] FIG. 3 shows a flowchart of the GRU network using multi-head attention mechanism and the improved fuzzy isolation forest algorithm in the embodiment shown in FIG. 1.

[0047] DETAILED DESCRIPTION OF THE INVENTION

[0048] The present application will be described in detail below in conjunction with the drawings and specific embodiments. Note that the aspects described below in conjunction with the drawings and specific embodiments are only exemplary and should not be understood as limiting the scope of protection of the present application in any way.

[0049] FIG. 1 shows a flowchart of an embodiment of a fault diagnosis method for a logistics mobile equipment based on digital twin of the present application. Please refer to FIG. 1, the implementation steps of the method of the present embodiment are described in detail as follows.

[0050] Step 1: Establish a corresponding digital twin model based on the logistics mobile equipment.

[0051] In step 1, the logistics mobile equipment is, for example, an AGV (Automated Guided Vehicle, also known as an automatic guided vehicle) equipment, and the model to be established is a digital twin model of the battery of the AGV equipment, including the following processing steps.

[0052] Firstly, Non-Uniform Rational B-Splines (NURBS) is introduced in the construction process of the digital twin model of AGV equipment to describe the contour shape of the AGV vehicle body. The specific processing steps are as follows: steps 1.1 to 1.4.

[0053] Step 1.1: Collect the contour point cloud data of the AGV vehicle body, and pre-process the point cloud data to remove noise and register the point cloud.

[0054] Step 1.2: Use the NURBS fitting algorithm to fit the pre-processed point cloud data to obtain the NURBS curve representation of the AGV vehicle body. This NURBS curve will approximately approximate the contour shape of the AGV vehicle body. Its mathematical expression is as follows:

[0055] P(u): a point on the NURBS curve, determined by the parameter u, is a coordinate point in three-dimensional space, representing a point on the vehicle body contour; n represents the maximum control vertex index.

[0056] N i , p(u): B-spline basis function, controls the shape of the curve, depends on the degree of the curve p and the position of the control vertex, here i is the index of the control vertex;

[0057] P i : the three-dimensional coordinates of the i-th control vertex, representing the control point of the vehicle body contour;

[0058] W i : the weight of the i-th control vertex, used to adjust the influence degree of the control vertex on the curve.

[0059] Step 1.3: In the NURBS fitting process, the position and value of the control vertex P i and the weight W i will affect the shape of the NURBS curve. By adjusting the position of the control vertex P i and the weight W i , the accuracy of curve fitting can be optimized, and key feature data can be preserved, so that the NURBS curve better matches the vehicle body contour. Specifically, a random sampling algorithm is used to adjust the position of the control vertex P i and the weight W i . First, set the sampling rate s, and randomly select a certain number of points from the original point cloud data as sampling points. The number of sampling points can be calculated by the sampling rate and the number of points in the original point cloud data, for example, the number of sampling points = sampling rate * number of points in the original point cloud data. The randomly selected sampling points are used as the down-sampled point cloud data for subsequent processing and fitting.

[0060] Suppose the original point cloud data has N points, and the coordinates of the points are represented as P i = [x i , y i , z i ], where i = 1, 2, …, N

[0061] Set the sampling rate: let the sampling rate be s, 0 < s <= 1.

[0062] Calculate the number of sampling points: the number of sampling points is M = s * N.

[0063] Randomly select sampling points: randomly select M points from the original point cloud data, and the indices of these points are I1, I2, …, I M .

[0064] Output the sampled point cloud: take the selected M points as the down-sampled point cloud data, that is, the output sampled point cloud is

[0065] {P{I1}, P{I2}, …, P{I M}}.

[0066] Step 1.4: Form the NURBS surface of the vehicle body surface by connecting the control vertices of the NURBS curve obtained in step 1.3, thereby realizing the 3D model reconstruction of the vehicle body.

[0067] Then, the processing steps of constructing the digital twin model of the battery of the AGV device include the following.

[0068] First, define the symbols and parameters involved in the digital twin model of the battery: considering the running state of the AGV, mainly taking the full load and empty load two states. In the empty load state, the battery current (I) is close to zero or very small, because the AGV device has no load. In the full load state, the battery current (I) is relatively large, because the AGV device bears additional load.

[0069] V: battery voltage;

[0070] I(empty): battery current in empty load state;

[0071] I(full): battery current in full load state;

[0072] R: battery internal resistance;

[0073] Q: battery charge state;

[0074] C: battery capacity;

[0075] T: battery temperature;

[0076] SOC: battery remaining capacity;

[0077] t: time.

[0078] First, construct the relationship function between battery open-circuit voltage and remaining capacity: The relationship function between battery open-circuit voltage and remaining capacity (SOC) and temperature (T) may differ in the unloaded and full-load states, as the internal chemical reactions and electrochemical state of the battery may differ under different load conditions. Therefore, the relationship function can be adjusted to two different functions:

[0079] AGV unloaded state: V empty (SOC, T) = V maxempty -a empty *SOC + b empty *T

[0080] AGV full-load state: V full (SOC, T) = V maxfull -a full *SOC + b full *T

[0081] where V empty represents the maximum open-circuit voltage of the battery in the unloaded state, V maxempty represents the maximum available voltage of the battery in the AGV unloaded state, V maxfull represents the maximum available voltage of the battery in the AGV full-load state

[0082] In the unloaded state, the open-circuit voltage of the battery is usually relatively high, as the internal chemical reactions and electrochemical state of the battery may be more inclined to maintain a higher voltage level without load.

[0083] a empty represents the slope of the change in open-circuit voltage of the battery in the unloaded state with respect to the remaining capacity (SOC). This parameter reflects the degree of influence of the change in SOC on the open-circuit voltage of the battery in the unloaded state. A larger a empty In the unloaded state, the open-circuit voltage of the battery is more likely to decrease or increase with the change in SOC.

[0084] b empty represents the slope of the change in open-circuit voltage of the battery in the unloaded state with respect to temperature (T). This parameter reflects the degree of influence of the change in temperature on the open-circuit voltage of the battery in the unloaded state. A larger b empty means that in the unloaded state, the open-circuit voltage of the battery is more likely to decrease or increase with the change in temperature.

[0085] Correspondingly, V fullV0represents the maximum open circuit voltage of the battery in full load state. In full load state, the internal chemical reaction and electrochemical state of the battery may cause the open circuit voltage to be relatively low due to the influence of bearing additional load.

[0086] a full V0represents the slope of the open circuit voltage of the battery in full load state with respect to the state of charge (SOC).

[0087] b full V0represents the slope of the open circuit voltage of the battery in full load state with respect to the temperature (T).

[0088] The specific values of these relationship function parameters can be obtained based on the battery specification table of the manufacturer.

[0089] Secondly, an improved function about the battery voltage dynamic model is constructed, which should now take into account the influence of battery temperature and state of charge:

[0090] AGV in empty state: V(t) = V0 empty (SOC(t), T(t)) - I empty *R-V soc (Q(t))

[0091] AGV in full load state: V(t) = V0 full (SOC(t), T(t)) - I full *R-V soc (Q(t))

[0092] V(t): the real-time voltage of the battery at time t.

[0093] V0 empty (SOC(t), T(t)): the open circuit voltage of the battery.

[0094] SOC(t): the state of charge of the battery at time t.

[0095] T(t): the temperature of the battery.

[0096] V0 soc (Q(t)) represents a correction factor for adjusting the open circuit voltage of the battery based on the state of charge of the battery at time t. Q(t) represents the remaining capacity of the battery at time t.

[0097] Then, based on the state of charge and temperature of the battery, a battery capacity decay model is constructed:

[0098] The battery capacity decay model and the battery temperature and SOC dynamic model are the same in empty and full load states, because these parts are mainly related to the capacity decay and SOC, temperature change of the battery, and are independent of the load of the battery. C(t) = C0 initial *fcapacity (SOC(t), T(t))*e -k*t

[0099] C(t): Represents the remaining capacity of the battery at time t, that is, the amount of charge the battery can still store at that moment. The battery capacity gradually decreases over time.

[0100] C initial : Indicates the initial capacity of the battery, that is, the amount of charge that the battery can store when it is fully charged at the beginning.

[0101] f capacity : This represents the capacity degradation factor that takes into account the effects of remaining charge (SOC) and temperature (T). This factor describes how the battery capacity decreases as SOC and temperature change.

[0102] k: This is the degradation coefficient, used to adjust the rate of battery capacity degradation. The larger the k value, the faster the battery capacity degrades; the smaller the k value, the slower the battery capacity degrades.

[0103] The dynamic model of battery temperature and SOC is as follows: The battery temperature and SOC also change with time and battery usage. The following differential equations are constructed to describe their dynamic evolution:

[0104] in, This represents the rate of change of battery temperature over time. t T(t) represents the capacity decay factor at time t; T(t) represents the battery temperature at time t; I(t) represents the battery current at time t; and V(t) represents the real-time voltage at time t.

[0105] This indicates the rate of change of battery SOC over time. The 3600 is used to convert the current unit to hours to match the SOC unit.

[0106] Step 2: Collect real-time data of the battery entity for adaptive model updates.

[0107] As shown in Figure 2, the refinement process in step 2 includes the following steps.

[0108] Step 2.1: Data collection. Install corresponding sensors in the AGV battery system to monitor the status of the battery in real time. For the empty state, temperature sensors, voltage sensors, and current sensors can be selected to collect data. For the full state, in addition to the above sensors, load sensors or weight sensors can be added to monitor the load state. The collection frequency is set in this embodiment. In the empty state, if the battery power is higher than 20%, the current, voltage, and temperature of the battery are collected at a frequency of 5s / time, and the remaining power is collected at a frequency of 1min / time. If the battery power is lower than 20%, the current, voltage, and temperature of the battery are collected at a frequency of 10s / time, and the remaining power is collected at a frequency of 2min / time. In the full state, if the battery power is higher than 20%, the current, voltage, and temperature of the battery are collected at a frequency of 10s / time, and the remaining power is collected at a frequency of 2min / time. If the battery power is lower than 20%, the current, voltage, and temperature of the battery are collected at a frequency of 15s / time, and the remaining power is collected at a frequency of 5min / time. The specific collection frequency can be adjusted according to the application scenario.

[0109] Step 2.2: Abnormal value detection preprocessing of collected data.

[0110] Due to the influence of technical level, connection condition, environment, etc., there will be some points with large deviation in the collected data, which are outliers. Abnormal values will reduce the accuracy of the prediction model and should be removed. This embodiment adopts a criterion method, named C-Letts criterion method.

[0111] The method is described as follows: calculate the residual error is the average value of the battery capacity attenuation from week 1 to week i, V i represents the battery voltage of the i-th week.

[0112] Calculate the standard deviation: n represents the number of data points, that is, the number of collected battery voltage values.

[0113] Calculate the deviation ratio:

[0114] Introduce adjustable parameters (a, b, c) to calculate variable A: A = a*max(|X i |), where a is the coefficient for amplifying the residual error.

[0115] Calculate variable B: where b is the coefficient for adjusting according to the average value of the residual error.

[0116] Calculate variable C: C = c*σ, where c is the coefficient for controlling the standard deviation.

[0117] Calculate the threshold value: T: The final threshold value is obtained by weighted combination of variables A, B, and C.

[0118] The adjustable parameter is used to control the sensitivity of the threshold value. i If the absolute value of the residual Xi corresponding to the measurement value Vi satisfies |Xi| > T, it is considered as an outlier and is removed. By introducing the deviation ratio and the adjustable parameter, this modified threshold setting formula can determine the threshold value according to the deviation degree of the data and the sensitivity requirement, making the outlier judgment more flexible and adjustable. The specific formula design and parameter selection should be adjusted according to the actual requirements and data characteristics.

[0119] Secondly, set the initial model evolution period as n, and determine the acceptable error range as E a , which represents the allowed relative error. In each period, calculate the relative error E

[0120] If the relative error E is within the acceptable range, i.e. E ≤ E a , it is considered that the prediction of the current model is relatively accurate, and the evolution period will remain or be extended to n+n / 2.

[0121] If the relative error E exceeds the acceptable range, i.e. E > E a , the model evolution period needs to be adjusted. At this time, the evolution period should be reduced to n / 2 to adapt to the changes in data and improve the accuracy of the model.

[0122] By dynamically adjusting the evolution period, the self-adaptive evolution method can flexibly change the evolution speed of the model according to the characteristics and error of the actual data, so as to better adapt to the changes in data and improve the accuracy of prediction. This can effectively deal with different stages of data characteristics and influence, improve the adaptability and prediction ability of the model.

[0123] Step 3: input the fault information into the bidirectional GRU model with multi-head attention mechanism.

[0124] As shown in Figure 3, the refinement processing steps of step 3 are as follows.

[0125] Step 3.1: input the collected AGV battery data in the digital twin model into the bidirectional GRU network, and the input data represents: the battery data sequence is X = {x1, x2,... x T}, where x t represents the battery data signal at time t.

[0126] Step 3.2: Forward GRU calculation, the forward GRU takes the input battery data sequence X and the hidden state at the previous time step, and calculates the hidden state at the current time step, here sigmoid activation function is adopted, the formula is as follows:

[0127] x represents the input battery data sequence, e represents the base of natural logarithm. The purpose of this function is to compress the input battery data value to a value between 0 and 1.

[0128] Step 3.3: Reverse GRU calculation, the reverse GRU takes the input battery data sequence X and the hidden state at the next time step, and calculates the hidden state at the current time step, also adopting sigmoid activation function.

[0129] Step 3.4: Weight to GRU hidden state. The calculation formula is as follows:

[0130] The hidden state weight represents the importance of the battery data hidden state of the forward GRU at time t; W a and b a respectively represent the weight matrix and the bias term; The hidden state of the forward GRU at time t represents the state information of the battery at time t, such as voltage, current, temperature; The normalized hidden state weight is calculated by the softmax function; The context vector represents the weighted sum of the hidden state of the forward GRU and its weight.

[0131] Step 3.5: Weighted sum of the extracted forward and reverse GRU hidden states, and output state information formula:

[0132] Where, z t is a scalar between 0 and 1, which is the weighted sum of the forward and reverse hidden states at time t, and the specific formula of z t is proposed as follows:

[0133] σ is the sigmoid activation function, W z is the learnable weight matrix, b z is the learnable bias term, The hidden state of the forward GRU at time t represents the state information of the battery at time t, such as voltage, current, temperature; The hidden state of the reverse GRU at time t represents the state information of the battery at time t, such as voltage, current, temperature.

[0134] Step 3.6: Use the multi-head attention mechanism to calculate the attention weight of the forward GRU hidden state, and then weight-sum with the context vector obtained by the multi-head attention mechanism to obtain the updated forward GRU hidden state. The calculation formula is as follows:

[0135] Context vector

[0136] Attention weight:

[0137] is the query, key and value of the hth head at time step t, respectively, represents the value of the feature representation related to the battery state, voltage, current, temperature of the hth head at time step t. Among them, softmax represents the softmax function for normalizing the attention weight.

[0138] Step 4: Combine the improved fuzzy isolation forest algorithm for anomaly point detection. Including the introduction of bivalent fuzzy theory in the fuzzy isolation forest algorithm and improvement.

[0139] The detailed processing steps of step 4 are as follows.

[0140] Step 4.1: For each feature F i in the data generated by the GRU network i , the membership function is represented by the envelope function as: F i (f) i

[0141] Where μF i (f) is the upper bound function of the membership of F i , and υF i (f) is the lower bound function of the membership of F i . In this way, each feature vector F i can express its fuzzy information with its envelope function, and these functions can be determined according to expert knowledge or experimental data.

[0142] Step 4.2: Prepare the training set data. Assume that the training set contains n samples, each with m features, denoted as

[0143] X = {x1, x1,... x n}, where x n = [x i1 , x i2 ,..., x im ] is the feature vector of the i-th sample.

[0144] Step 4.3: Build an isolation tree, randomly select a sample subset from the training set, denoted as X​s where |X s |≤n. For each isolated tree T i , the tree structure is constructed in a recursive way.

[0145] Step 4.4: Partition the data, for the current node N, randomly select a feature q and a split point p, divide the sample set Xninto left subset X left and right subset X right . Mark the current node N as a leaf node. Divide the current node N into two child nodes, and recursively construct the left sub-tree T left and right sub-tree T right , here the termination condition is set as when there is only one sample in the subset, then terminate.

[0146] Step 4.5: Calculate the average path length, the formula is as follows:

[0147] where H(k) is the harmonic number, which can be represented by an estimated value: H(k) = ln(k) + γ, where γ is the Euler constant, about 0.5772156649.

[0148] Step 4.6: Calculate the anomaly score, by calculating the average of the normalized path length, get the final anomaly score:

[0149] where E(h(x, T i )) is the anomaly score function, E is used to quantify the degree of anomaly of the data point; h(x, T i ) represents the battery data hidden state at time step Ti obtained by the data path x.

[0150] Step 4.7: Calculate the two-type fuzzy membership degree, according to the anomaly score and the average path length to calculate the μF i (f) membership degree upper bound, υF i (f) membership degree lower bound, the formula is:

[0151] θ and θ' are parameters that control the shape of the membership degree upper and lower bound curves. The obtained upper and lower bounds are used to calculate the final membership degree of each test sample. c(n) represents the average path length; f represents the value of each feature F i in the data generated by the GRU network, which is used as an input variable here.

[0152] Step 4.8: Determine the factor set U of the evaluation object. U = {u1, u2,..., u m}, determine the evaluation set V, V = { normal, abnormal, suspicious, unknown}. Perform single factor evaluation: calculate the membership degree of each evaluation factor and its corresponding evaluation result to obtain the membership value of single factor. The fuzzy relation matrix R is represented as follows:

[0153] wherein, r ij represents the membership value of the evaluation factor μ i corresponding to the evaluation result υ i .

[0154] Step 4.9: Fuzzy comprehensive evaluation result analysis. Arrange the evaluation results in order from good to bad, and use rank to represent, rank starts from 1 and increases. Then, by the method of weighted summation, the final relative position A of the object to be evaluated is obtained:

[0155] wherein, k is a pending coefficient, used to prevent some evaluation results from having too much influence on the results, wherein the evaluation object with a larger relative position is regarded as an abnormal point, represents the fuzzy comprehensive evaluation result of the jth object to be evaluated under the pending coefficient k.

[0156] Step 5: Use the obtained fault data to implement the solution measures according to the risk level.

[0157] The use of the obtained fault data in step 5 is obtained by the method of analytic hierarchy process, and the detailed processing steps of the risk level are as follows.

[0158] Step 5.1: Divide the hierarchy. Level 1: risk level Level 2: equipment (AGV, stacker, conveyor) Level 3: fault type (battery failure, conveyor wear, stacker motor failure).

[0159] Step 5.2: For each level and factor in the hierarchy structure, construct a judgment matrix, wherein each element represents the comparison between two factors. Let A represent the judgment matrix, and A(i, j) represent the importance score of the ith factor relative to the jth factor. The judgment matrix is a positive reciprocal matrix, which satisfies the following conditions:

[0160] A(i, j) > 0: indicates the importance score of the ith factor relative to the jth factor.

[0161] A(i, j) = 1 / A(j, i): indicates the symmetry of the judgment matrix.

[0162] A(i, j)*A(j, k) = A(i, k): indicates the transitivity of the judgment matrix.

[0163] The specific judgment matrix is shown as follows:

[0164] where A, B, C, D, E, F are the importance scores of relative comparison, representing the importance comparison between the device and the fault type.

[0165] Step 5.3: Calculate the feature vector. Let w represent the feature vector, and w(i) represent the weight of the i-th factor. Normalize the feature vector, and for each factor of each fault type, assign it a corresponding score. Let x(i, j) represent the score of the i-th factor of the j-th fault type.

[0166] Step 5.4: Calculate the risk level. For each fault type, multiply its corresponding factor score by the corresponding normalized weight and perform a weighted sum to obtain the risk level of the fault type. Let R represent the risk level, R(j) represent the risk level of the j-th fault type, and x represent the factor score vector. The calculation formula is as follows: R(j) = ∑(w(i)*x(i, j))

[0167] Specifically, for each fault type, there can be multiple factor scores and corresponding normalized weights, and the risk level can be obtained using the weighted sum method. For example, assuming there are n factor scores and normalized weight corresponding values x1, x2,..., xn and w1, w2,..., wn, the calculation formula of the risk level is as follows: R(j) = w1*x1(j) + w2*x2(j) +... + wn*xn(j),

[0168] where x1(j), x2(j),..., xn(j) represent the factor scores of the j-th fault type, and w1, w2,..., wn represent the normalized weight vector.

[0169] The specific steps of implementing the solution measures according to the risk level in step 5 are as follows.

[0170] Automatic Guided Vehicle (AGV): Low-risk fault: sensor aging or uncalibration leading to insufficient sensitivity. Solution: Regularly calibrate the sensor to ensure its accuracy. Medium-risk fault: battery aging or not charging. Solution: Implement regular maintenance and charging plan, replace the aging battery. High-risk fault: Cause: motor abnormality leading to device overheating. Solution: Monitor the motor temperature, and cool down or stop cooling in time.

[0171] Conveyor: Low risk failure: Drive belt damage due to aging, impact from goods. Solution: Regularly check and adjust tension of drive belt. Medium risk failure: Motor control system failure due to excessive current load. Solution: Backup control system, replace faulty controller in time. High risk failure: Sensor aging or abnormal signal due to uncalibration. Solution: Regularly check and calibrate sensors to ensure their accuracy.

[0172] Stacker crane: Low risk failure: Wear and tear of crane mechanical parts. Solution: Regular lubrication and replacement of worn mechanical parts. Medium risk failure: Communication system failure. Solution: Backup communication system, repair or replace faulty equipment in time.

[0173] High risk failure: Stacker crane control system failure. Solution: Backup control system, regular backup and update of control program.

[0174] Picking robot: Low risk failure: Loose mechanical joints. Solution: Regularly check and tighten mechanical joints.

[0175] Medium risk failure: Cause: Vision system failure. Solution: Regular calibration and maintenance of vision system, check camera status and replace damaged parts. High risk failure: Cause: Control system failure or program error. Solution: Backup control system, regular update and testing of control program.

[0176] Although the above methods are illustrated and described as a series of acts, it will be appreciated that not all of the acts described are necessary to implement the methods in accordance with one or more embodiments and that implen enting the described methods can involve additional or different acts in some cases.

[0177] Those skilled in the art will further appreciate that the various illustrative logical blocks, modules, circuits, and algorithm steps described in connection with the embodiments disclosed herein can be implemented as electronic hardware, computer software, or combinations of both. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps have been described above generally in terms of their functionality, without reference to a particular sequence of acts. Such functionality can be implemented in a variety of ways, and in one embodiment, the described functionality can be implemented by one or more computer software programs running on a general purpose computer. In another embodiment, the described functionality can be implemented by one or more hardware logic circuits, such as a

[0178] The various illustrative logical blocks, modules, and circuits described in connection with the embodiments disclosed herein can be implemented or performed with a general purpose processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. A general purpose processor can be a microprocessor, but in the alternative, the processor can be any conventional processor, controller, microcontroller, or state machine. A processor can also be implemented as a combination of computing devices, e.g., a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP core, or any other such configuration.

[0179] The steps of a method or algorithm described in connection with the embodiments disclosed herein can be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. A software module can reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor such that the processor can read information from, and write information to, the storage medium. In the alternative, the storage medium can be integral to the processor. The processor and the storage medium can reside in an ASIC. The ASIC can reside in a user terminal. In the alternative, the processor and the storage medium can reside as discrete components in a user terminal.

[0180] In one or more exemplary embodiments, the functions described can be implemented in hardware, software, firmware, or any combination thereof. If implemented in software as a computer program product, the functions can be stored on or transmitted over as one or more instructions or code on a computer-readable medium. Computer-readable media includes both computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another. A storage media can be any available media that can be accessed by a computer. By way of example, and not limitation, such computer-readable media can comprise RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer. Also, any connection is properly termed a computer-readable medium. For example, if the software is transmitted from a website, server, or other remote source using a coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of medium. Disk and disc, as used herein, includes compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), floppy disk and blu-ray disc where disks usually reproduce data magnetically, while discs reproduce data optically with lasers. Combinations of the above should also be included within the scope of computer-readable media.

[0181] The previous description of the disclosure is provided to enable any person skilled in the art to make or use the disclosure. Various modifications to the disclosure will be readily apparent to those skilled in the art, and the generic principles defined herein can be applied to other variations without departing from the spirit or scope of the disclosure. Thus, the disclosure is not intended to be limited to the examples described herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1.A method for diagnosing a fault of a logistics mobile device based on digital twinning, characterized in that, The method comprises: Step 1: establishing a corresponding digital twin model based on the logistics mobile device; Step 2: collecting real-time data of the battery entity for model adaptive updating; Step 3: inputting fault information into a bidirectional GRU model with a multi-head attention mechanism; Step 4: combining an improved fuzzy isolation forest algorithm for anomaly point detection, including introducing bivalent fuzzy theory and improving in the fuzzy isolation forest algorithm; Step 5: using the obtained fault data to implement solving measures according to the risk level. 2.The digital-twin-based fault diagnosis method of a logistics mobile device according to claim 1, wherein, The model to be established in step 1 is a digital twin model of the battery of the AGV device, including the following processing steps: Step 1.1: collecting contour point cloud data of the AGV vehicle body, preprocessing the point cloud data to remove noise and registering the point cloud; Step 1.2: using NURBS fitting algorithm to curve fit the preprocessed point cloud data to obtain NURBS curve representation of the AGV vehicle body; Step 1.3: controlling the position and value of the control points and weights to affect the shape of the NURBS curve during NURBS fitting; Step 1.4: connecting the control points of the NURBS curve obtained in step 1.3 to form a NURBS surface of the vehicle body surface, thereby realizing 3D model reconstruction of the vehicle body. 3.The digital-twin-based fault diagnosis method of a logistics mobile device according to claim 1, wherein, Step 2 further comprises: Step 2.1: data collection; Step 2.2: abnormal value detection preprocessing of the collected data. 4.The method of claim 1, wherein, Step 3 further comprises: Step 3.1: inputting the collected AGV battery data in the digital twin model into the bidirectional GRU network; Step 3.2: forward GRU calculation; Step 3.3: reverse GRU calculation; Step 3.4: weight to GRU hidden state; Step 3.5: weighted sum of the extracted forward and reverse GRU hidden states, and output state information formula; Step 3.6: using a multi-head attention mechanism to calculate the attention weight of the forward GRU hidden state, and then weighting and summing the context vector obtained through the multi-head attention mechanism to obtain the updated forward GRU hidden state. 5.The digital-twin-based fault diagnosis method of a logistics mobile device according to claim 1, wherein, Step 4 further comprises: Step 4.1: expressing the membership function with an envelope function for each feature in the data generated by the GRU network; Step 4.2: preparing training set data; Step 4.4: dividing data; Step 4.5: calculating the average path length; Step 4.6: calculating the anomaly score, obtaining the final anomaly score by calculating the average value of the normalized path length; Step 4.7: calculating the bivalent fuzzy membership, calculating the upper membership bound and lower membership bound of each test sample according to the anomaly score and the average path length; Step 4.8: determining the factor set of the evaluation object; Step 4.9: fuzzy comprehensive evaluation result analysis. 6.The logistics mobile equipment fault diagnosis method based on digital twinning according to claim 1, wherein, Step 5 further comprises: Step 5.1: dividing levels; Step 5.2: constructing a judgment matrix for each level and factor in the hierarchical structure; Step 5.3: calculating the characteristic vector; Step 5.4: calculating the risk level.

Images