Sensor automatic detection method and system based on supply chain

Abstract

The invention relates to the technical field of supply chain monitoring, and particularly discloses an automatic sensor detection method and system based on a supply chain, and the method comprises the steps: generating a hierarchical pressure gradient field and an event node distribution diagram through obtaining the transportation vibration energy; if a creep coefficient resetting algorithm is triggered, mapping the event node graph into a weighted directed graph, extracting a key conduction path, and positioning an unpacking release type drift node; and introducing an unpacking stress correction term expansion cross-correlation entropy model, and determining a priority compensation strategy in combination with a time window information entropy to realize dynamic correction of sensor drift. The system comprises a hierarchy analysis module, a topology construction module, a lag positioning module and a drift compensation module. Through cross-scale modeling of vibration energy, material creep and drift compensation, the problem of sensor reading lag caused by stress accumulation in a supply chain can be reduced, and the packaging optimization decision efficiency is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of supply chain monitoring, in particular to a sensor automatic detection method and system based on a supply chain. BACKGROUND

[0002] In the monitoring of supply chain cargo transportation, sensors are the key equipment for obtaining cargo state information, but the existing technology has significant defects:

[0003] When the cargo package is opened, the instantaneous release of the long-term accumulated compression stress will cause the deformation of the sensor installation position, resulting in a lagging drift in the reading. The existing scheme fails to effectively distinguish this unpacking release type drift from the conventional drift caused by transportation vibration, making it difficult to locate the drift source, often resulting in misjudgment or missed judgment, affecting the reliability of sensor data, and further hindering accurate monitoring of the state of goods in the supply chain and optimization decisions for packaging processes.

[0004] Therefore, the present application provides a sensor automatic detection method and system based on a supply chain. SUMMARY

[0005] The present application aims to provide a sensor automatic detection method and system based on a supply chain to solve the above background problems.

[0006] The object of the present application can be achieved by the following technical solutions:

[0007] A sensor automatic detection method based on a supply chain, comprising the following steps:

[0008] Obtaining the transportation vibration energy of a supply chain edge vibration sensor, performing gradient analysis on the transportation vibration energy to obtain a hierarchical pressure gradient, and establishing a node fingerprint vector and an event node distribution map based on the hierarchical pressure gradient;

[0009] By sensing the package opening and closing signal and the strain data, creep coefficient reset trigger analysis is performed. If the reset is triggered, the event node distribution map is mapped to a weighted directed graph. The node transmission factor and path score of the weighted directed graph are extracted in combination with the material creep accumulation effect to construct a risk potential model and output a path conduction topology graph;

[0010] Based on the path conduction topology graph, a drift analysis path is extracted, the vibration drift slope and the release drift slope of the stress after unpacking in the drift analysis path are calculated, and the creep sensitive area risk nodes with significant lagging effect are located;

[0011] Based on the unpacking stress correction model of the creep risk node mutual correlation entropy, the mutual correlation entropy of the appearance probability of the time window strong correlation node group is combined to screen the priority compensation time window and develop a compensation strategy for the drift sensor.

[0012] As a further scheme of the present application, the manner of transporting vibration energy for the gradient analysis is:

[0013] A three-dimensional pressure distribution field matrix is constructed by acquiring the pressure load sensitive to the goods, a gradient vector of the pressure distribution field is extracted by a central difference method, and a hierarchical pressure gradient field is constructed;

[0014] Different feature sampling points are set for the hierarchical pressure gradient field, texture features and level features of the sampling points are extracted, and the texture features and level features of the sampling points are processed according to the event principle;

[0015] The sampling points meeting the event principle are marked as event nodes, the texture features and level features of the event nodes are processed by space-time coding and a node fingerprint vector is constructed;

[0016] All node fingerprint vectors of the goods at different stages of the supply chain are acquired, and an event node distribution map is established according to a node rule.

[0017] As a further scheme of the present application, the manner of acquiring the pressure load sensitive to the goods is:

[0018] Vibration acceleration collected by an edge vibration sensor of a supply chain is acquired, a hierarchical energy ratio is extracted by multi-layer decomposition of the vibration acceleration signal through discrete wavelet transform;

[0019] A stacked energy conversion model is constructed to convert the hierarchical energy ratio into pressure loads of different energy levels of a packaging contact surface during transportation of goods of the supply chain;

[0020] The pressure loads of different levels are processed to screen and obtain the pressure load sensitive to the goods.

[0021] As a further scheme of the present application, the manner of performing the creep coefficient reset trigger analysis is:

[0022] A packaging opening and closing signal is acquired, and an opening and closing state of the packaging is determined based on the packaging opening and closing signal;

[0023] If the packaging is in the opening and closing state, a creep coefficient reset algorithm is triggered to reset the creep coefficient to obtain a real-time creep coefficient.

[0024] As a further scheme of the present application, the manner of constructing the risk potential model output path conduction topology graph is:

[0025] A weighted directed graph and a node fingerprint vector of the weighted directed graph are acquired, a node transmission factor and a path score are extracted in combination with a material creep cumulative effect;

[0026] A risk potential model is constructed based on the node transmission factor and the path score, and the key path of the conduction efficiency and the creep risk is extracted through the risk potential model;

[0027] Based on the obtained key path, a path conduction topology graph is output to the upstream and downstream of the supply chain.

[0028] As a further scheme of the present application, the manner of obtaining the weighted directed graph is:

[0029] A packaging state vector is constructed by obtaining a real-time creep coefficient, a stress release coefficient and a packaging opening and closing state;

[0030] The time stamp of the packaging state vector and the time stamp of the node fingerprint vector are obtained, and the packaging state vector and the node fingerprint vector are aligned and spliced based on the time stamp to establish a node feature vector;

[0031] The weight between any two event nodes in the event node distribution graph is collected, and the event node distribution graph is mapped into a weighted directed graph.

[0032] As a further scheme of the present application, the extraction manner of the weight between the any two nodes is:

[0033] Any two event nodes in the event node distribution graph are obtained, if the two event nodes are in the same isochronous node group, the Euclidean distance of the node feature vectors of the two event nodes is calculated as a node judgment distance;

[0034] If the two event nodes are not in the same isochronous node group, or the node judgment distance is higher than or equal to a preset node judgment distance threshold, a node stress conduction analysis is directly performed to obtain a pressure gradient difference;

[0035] The node feature vectors and the time stamps of any two nodes are calculated in space-time correlation to obtain a space-time correlation degree;

[0036] A weight equation is constructed, the pressure gradient difference, the space-time correlation degree and the real-time creep coefficient are input into the weight equation to obtain the weight between any two nodes, and the weight between any two nodes is obtained.

[0037] As a further scheme of the present application, the manner of positioning the creep sensitive area risk node with significant hysteresis effect is:

[0038] A path with maximum potential accumulation and continuously increasing creep coefficient is extracted from the path conduction topology graph as a drift analysis path;

[0039] The pressure gradient value of each event node in the drift analysis path in a time window is obtained, the change rate of the pressure gradient value in the time window is calculated, and a vibration drift slope is obtained;

[0040] Obtaining the stress release drift slope in the detection time window after the package is in the open and closed state;

[0041] Determining the drift type based on the vibration drift slope and the release drift slope, if it is a release type drift, obtaining the hysteresis influence factor and judging whether the hysteresis effect is significant;

[0042] If the hysteresis effect is significant, setting a creep sensitive region based on an event node distribution map, and a creep risk node in the creep sensitive region in the drift analysis path.

[0043] As a further scheme of the application: the way of screening the priority compensation time window is:

[0044] Obtaining the node feature vectors of any two creep risk nodes, calculating the cross-correlation entropy of the node feature vectors of the two creep risk nodes, and extending and correcting the cross-correlation entropy to obtain the corrected cross-correlation entropy;

[0045] Based on the cross-correlation entropy of any two creep risk nodes, entropy value constraint modeling of the cooperative compensation range is carried out to obtain a strong correlation node group sensitive to unpacking;

[0046] Obtaining the probability of different strong correlation node groups appearing in each time window in the supply chain process, performing information entropy value analysis on the time window to obtain time node combination entropy,

[0047] Based on the time node combination entropy, the priority compensation time window is determined.

[0048] A supply chain-based sensor automatic detection system: comprising the following modules:

[0049] The hierarchical analysis module is used to obtain the transportation vibration energy of the supply chain edge vibration sensor, perform gradient analysis on the transportation vibration energy to obtain a hierarchical pressure gradient, and establish a node fingerprint vector and an event node distribution map based on the hierarchical pressure gradient;

[0050] The topology construction module is used to perform creep coefficient reset trigger analysis through sensing the package opening and closing signal and the strain data, if the reset is triggered, the event node distribution map is mapped into a weighted directed graph, the node transmission factor and the path score of the weighted directed graph are extracted in combination with the material creep cumulative effect, and a risk potential model is constructed to output a path conduction topology graph;

[0051] The hysteresis positioning module is used to extract a drift analysis path based on the path conduction topology graph, calculate the vibration drift slope and the stress release drift slope in the drift analysis path, and locate the creep sensitive region risk node with significant hysteresis effect;

[0052] Drift compensation module: based on the creep risk node cross-correlation entropy of the unpacking stress correction model, for combining the cross-correlation entropy of the time window strong correlation node group appearance probability, screen priority compensation time window and make the compensation strategy of drift sensor.

[0053] The beneficial effects of the present application are:

[0054] (1) Quantitative mapping of the transport vibration energy to the stress at the packaging interface, the constructed space-time coupled hierarchical pressure gradient field can capture the pressure distribution characteristics of different energy levels. Based on the event node distribution map generated by local binary pattern and space-time coding, the space-time aggregation mode of pressure anomaly of each link of the supply chain is presented, which provides basic data support with physical meaning and space-time dimension for subsequent risk transmission analysis, and improves the identification accuracy and positioning efficiency of abnormal events;

[0055] (2) Use micro-electro-mechanical system sensor to sense the opening and closing state of the package, trigger the dynamic reset algorithm of the creep coefficient, solve the problem of material parameter mutation caused by stress release after unpacking. Construct a weighted directed graph and extract high-risk transmission path, realize the upgrade from single-point vibration analysis to supply chain network level risk prediction. It is beneficial to simultaneously optimize the packaging buffer design and transportation strategy, and effectively improve the accuracy of risk path identification and packaging process decision efficiency.

[0056] (3) By segmenting the vibration drift slope and the unpacking stress release slope, combining the material creep theory to distinguish the drift type, and positioning the sensor reading lag problem caused by unpacking stress release. Based on the BURGERS model parameter analysis of creep lag effect, the sensor drift trend caused by material aging can be predicted in advance, which provides a quantitative basis for the time dimension of packaging material durability evaluation and risk warning, and is beneficial to enhance the reliability of sensor data;

[0057] (4) Introducing the unpacking stress correction term to extend the cross-correlation entropy model, combining the time window information entropy to analyze the dynamic compensation opportunity, realizing the decision of drift compensation, and the differential compensation strategy generated by multi-parameter fusion can be used to adjust the packaging structure and supply chain execution strategy, forming a closed-loop optimization system of monitoring and compensation, effectively improving the anti-risk ability of the supply chain and the self-adaptive adjustment performance of the packaging system. BRIEF DESCRIPTION OF DRAWINGS

[0058] The present application will be further described below in conjunction with the drawings.

[0059] Figure 1 is a flow chart of a sensor automatic detection method based on a supply chain according to the present application;

[0060] Figure 2 is a judgment flow chart for node stress transmission analysis in the present application;

[0061] Figure 3is a module diagram of a supply chain-based sensor automatic detection system. DETAILED DESCRIPTION

[0062] 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. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work are within the protection scope of the present application.

[0063] Embodiment 1

[0064] Please refer to Figure 1 The present application is a supply chain-based sensor automatic detection method, which comprises the following steps:

[0065] Step 1: obtaining the transportation vibration energy of the supply chain edge vibration sensor, performing gradient analysis on the transportation vibration energy to obtain a hierarchical pressure gradient, and establishing a node fingerprint vector and an event node distribution map based on the hierarchical pressure gradient;

[0066] The way of obtaining the transportation vibration energy of the supply chain edge vibration sensor and performing gradient analysis on the transportation vibration energy is as follows:

[0067] The vibration acceleration collected by the supply chain edge vibration sensor is obtained, the vibration acceleration signal is decomposed in multiple layers through discrete wavelet transform, and the energy ratio of each layer is extracted to obtain a hierarchical energy ratio;

[0068] A stacked energy conversion model is constructed to convert the hierarchical energy ratio into pressure load of different energy levels of the packaging contact surface during the transportation of the supply chain goods;

[0069] A person skilled in the art can understand that the three-axis acceleration signal collected by the edge vibration sensor is decomposed in 3 layers through db4 wavelet basis of discrete wavelet transform (DWT), and the signal is separated into high-frequency detail component (100-200Hz), medium-frequency detail component (50-100Hz) and low-frequency approximation component (0-50Hz);

[0070] The energy ratio is calculated by the sum of the coefficients of each layer, and a hierarchical energy ratio vector E is obtained l =[E1, E2, E3]; when constructing the stacked energy conversion model, first, based on the dynamic equation (m is the mass of the goods, Δt0 is the sampling interval, and l is the number of the hierarchical level) the energy ratio of each layer is inversely calculated as the acceleration peak value a lIn combination with the three-dimensional grid model of the package contact surface (the contact area is obtained through CAD), the peak acceleration is multiplied by the instruction to obtain dynamic pressure, the dynamic pressure is divided by the area of the contact point to obtain the dynamic pressure load of each contact point, and the damping coefficient of the packaging material is introduced to correct the pressure to form a three-dimensional pressure load matrix containing different energy levels, thereby realizing the quantitative mapping of vibration energy to the stress of the package interface.

[0071] The pressure loads of different levels are screened to obtain the pressure loads of the sensitive level of the goods.

[0072] A three-dimensional pressure distribution field matrix is constructed based on the pressure loads sensitive to the goods, the gradient vector of the pressure distribution field is extracted through the central difference method, and a hierarchical pressure gradient field is constructed.

[0073] It should be explained that based on the three-level pressure load, the goods fragility curve (such as glass products sensitive to high frequency) is obtained, the energy proportion threshold is set, and the sensitive level load is screened out; secondly, the node pressure values of the three-dimensional packaging grid in the CAD model are arranged according to the time stamp to construct a dynamic matrix of T x M x N x P (T is the time frame, and M / N / P is the spatial dimension); the central difference method is used for each spatial node to calculate the gradient of the node pressure value in the M, N and P directions, and the gradient vector field is combined to form a time-space coupled hierarchical pressure gradient field through Gaussian filtering to smooth the noise;

[0074] Among them, the way to establish the node fingerprint vector and the event node distribution map is:

[0075] Different feature sampling points are set for the hierarchical pressure gradient field, the texture features and level features of the sampling points are extracted, and the texture features and level features of the sampling points are event screened according to the event principle;

[0076] Among them, the texture features include: the 3x3 neighborhood pressure gradient values of the sampling points in the hierarchical pressure gradient field, which are finally compressed into 16-dimensional histograms;

[0077] The level features include: normalizing the absolute value of the pressure gradient to 0-5 levels, corresponding to the material fragility curve, for example, the sensitive area of electronic components is greater than or equal to 3 levels to trigger a warning;

[0078] The sampling points meeting the event principle are marked as event nodes, and the texture features and level features of the event nodes are spatio-temporally coded and the node fingerprint vector is constructed;

[0079] All node fingerprint vectors of the goods at different stages of the supply chain are obtained, and an event node distribution map is established according to the node rules.

[0080] Among them, the node rules include:

[0081] The node rule one is that if the construction times of the plurality of node fingerprint vectors are same, the node fingerprint vectors with the same time are classified into a group to obtain an isochronous node group;

[0082] The node rule two is that a position and a time of the isochronous node group in the supply chain are obtained, and a node distribution map is constructed.

[0083] It can be understood that the sampling point marked as an event node is screened through the event principle, the texture feature and the level feature are extracted, and the node fingerprint vector is generated by combining the space-time information coding; then all node fingerprint vectors of the goods in each stage of the supply chain are collected, and the node rule is processed: first, the fingerprint vectors are clustered according to the construction time, and the fingerprint vectors with the same time stamp are classified into a group to form an isochronous node group; then the specific position (such as a storage area, a transportation section, etc.) and the common time information of each isochronous node group in the supply chain are extracted, and the position and time data are mapped through visualization or a graph structure tool, and finally an event node distribution map is constructed, and the distribution state of different nodes at the same time point in the supply chain is intuitively presented.

[0084] The purpose of constructing the event node distribution map is:

[0085] The first purpose is to visually present the spatial distribution state of each event node in the supply chain at the same time dimension, so as to quickly identify abnormal nodes and provide intuitive basis for real-time monitoring and abnormal early warning of the supply chain.

[0086] The second purpose is to analyze the distribution rule and evolution track of the isochronous node group at different time points, and to mine the node correlation mode of each link of the supply chain, such as the aggregation characteristics of the isochronous node group at a certain stage, so as to provide data support for optimizing the resource allocation of the supply chain and improving the process efficiency.

[0087] The second step is to analyze the creep coefficient reset trigger through the sensing of the package opening and closing signal and the strain data, if the reset is triggered, the event node distribution map is mapped into a weighted directed graph, the node transmission factor and the path score of the weighted directed graph are extracted by combining the material creep cumulative effect, and a risk potential model is constructed to output a path conduction topology map.

[0088] The way of analyzing the creep coefficient reset trigger is:

[0089] Preferably, the MEMS (micro-electro-mechanical system) sensor at the sealing position of the package real-time senses the package opening and closing signal S(t) and the strain data of the package material, and synchronizes the package opening and closing signal and the strain data of the package material with the time stamp of the edge vibration sensor.

[0090] t is the trigger time of the package opening and closing signal S(t);

[0091] The strain data includes a stress release coefficient, stress change data of the packaging material during loading and unloading is collected in real time based on the stress sensor, and stress peak values and residual stress values in different stages are recorded; the basic coefficient is calculated by the ratio of the difference between the measured peak stress and the residual stress and the initial stress in combination with the elastic parameters of the packaging material and the dynamic load characteristics (such as vibration frequency and impact force) in the transportation process; finally, the stress release coefficient is corrected according to the coefficient fluctuation under different road conditions and loads in multiple batches of transportation tests, so as to obtain the stress release coefficient that fits the actual transportation working conditions;

[0092] Based on the packaging opening and closing signal S(t), the packaging opening and closing signal S(t) = 1, that is, the packaging is in an open state, and vice versa, the packaging is in a closed state;

[0093] If the packaging is in an open state, the creep coefficient resetting algorithm is triggered to reset the creep coefficient;

[0094] Preferably, the creep coefficient is reset by the formula: C r (t) = C r (t) old *(1-λ*Δσ), to obtain the real-time creep coefficient C r (t).

[0095] Wherein, C r (t) old is the creep coefficient before resetting, and λ and Δσ are the stress release coefficient and the average stress value of the packaging contact surface in the stress distribution field, respectively;

[0096] The real-time creep coefficient, the stress release coefficient and the packaging opening and closing state are obtained to construct a packaging state vector;

[0097] The time stamp of the packaging state vector and the time stamp of the node fingerprint vector are obtained, and the packaging state vector and the node fingerprint vector are aligned and spliced based on the time stamp;

[0098] Based on the aligned and spliced packaging state vector and node fingerprint vector, a node feature vector is established;

[0099] The weight between any two event nodes in the event node distribution diagram is collected, and the event node distribution diagram is mapped into a weighted directed graph;

[0100] Wherein, the extraction method of the weight between any two nodes is:

[0101] Any two event nodes in the event node distribution diagram are obtained, if the two event nodes are in the same isochronous node group, the Euclidean distance of the node feature vectors of the two event nodes is calculated as the node judgment distance;

[0102] If the two event nodes are not in the same isochronous node group, directly conduct node stress conduction analysis;

[0103] As shown in Figure 2 , the node judgment distance is compared with the preset node judgment distance threshold value, and if the node judgment distance is higher than or equal to the preset node judgment distance threshold value, node stress conduction analysis is conducted;

[0104] If the node judgment distance is lower than the node judgment distance threshold value, the change of the node judgment distance is continuously monitored;

[0105] The way of conducting node stress conduction analysis is:

[0106] The gradient vectors of the two nodes in the pressure distribution field are obtained, and the pressure gradient difference of any two nodes is calculated by the Euclidean distance;

[0107] It can be understood that the node stress conduction analysis can be used to quantify the stress conduction potential between nodes;

[0108] The node feature vectors of any two nodes and the timestamps corresponding to the two nodes are obtained, and the spatiotemporal correlation degree Att(i, j) is calculated by a multi-head attention network algorithm for spatiotemporal correlation calculation of the node feature vectors and the timestamps of any two nodes;

[0109] Those skilled in the art can understand that the implementation process of calculating the spatiotemporal correlation degree by the multi-head attention network is as follows: for any two nodes, first, fuse the node feature vectors (including physical properties, state parameters, etc.) and the time-stamped features (including time periodicity encoding vectors and relative time differences) processed by time embedding into input features; then, use N-head attention mechanism, take the fusion features of one node as the query vector, the other node as the key and value vectors, calculate the attention weight of each head by the scaling dot product formula, and aggregate the intermediate features to obtain the spatiotemporal correlation degree in the range of 0-1 after normalization;

[0110] The weight equation is: The weight w of the directed graph is obtained ij ;

[0111] Wherein, α, β are preset dynamic coefficients, C r (t) is the real-time creep coefficient, t is the time, and Δt is the time interval corresponding to the event nodes i and j;

[0112] P ij is the pressure gradient difference between the event nodes i and j;

[0113] The way of extracting the node transmission factor and path score of the weighted directed graph in combination with the material creep cumulative effect is:

[0114] Based on the node fingerprint vector of the weighted directed graph, the node transmission factor and the path score are extracted in combination with the cumulative effect of material creep;

[0115] Based on the node transmission factor and the path score, a risk potential model is constructed, and the key path of the conduction efficiency and the creep risk is extracted through the risk potential model;

[0116] Based on the obtained key path, a path conduction topology graph is output to the upstream and downstream of the supply chain;

[0117] The node transmission factor is obtained in the following manner:

[0118] The node transmission factor E is extracted through formula one: i ;

[0119] Wherein, Pr(i) is the level feature of event node i, and τ is the integral variable;

[0120] The path score is obtained through formula two:

[0121] It can be understood that a three-dimensional potential field is constructed based on the node transmission factor and the path score, the node potential is the node transmission factor E i , and the edge potential is the weight w ij of the directed graph;

[0122] Through a dynamic programming algorithm, a path with the maximum potential accumulation and the continuously increasing creep coefficient is searched starting from the warehouse and transportation stages: preferentially selecting edges with a gradient difference > 5kPa / cm and a time-space correlation degree > 0.8, and pruning low-creep paths with C_r(t) < 0.6;

[0123] The path conduction topology graph is constructed to achieve the following effects:

[0124] Effect one: the path conduction logic and the correlation of each event node in the supply chain, directly showing the transfer link of the goods state, information or risk between nodes, helping to quickly sort out the conduction main line and branch structure in the whole process

[0125] Effect two: by analyzing the connection strength and conduction path length of the nodes in the topology graph, the conduction efficiency of each link is quantitatively evaluated, and the redundant nodes or blocked links in the conduction process are located, providing a basis for optimizing path design and improving overall conduction efficiency;

[0126] Effect three: used for simulating the conduction influence range when a node is abnormal, predicting the diffusion path and the extent of the risk in the path, and providing support for formulating targeted risk blocking strategies.

[0127] Embodiment 2

[0128] Please refer to​​Figure 1 As shown, the application is a sensor automatic detection method based on a supply chain, further comprising the following steps:

[0129] Step three, based on the path conduction topology graph, extract the drift analysis path, calculate the vibration drift slope and the stress release drift slope after unpacking in the drift analysis path, and locate the creep sensitive area risk node with significant hysteresis effect;

[0130] Preferably, the path with the maximum potential energy accumulation and the continuous growth of the creep coefficient is extracted from the path conduction topology graph as the drift analysis path;

[0131] Get the stress gradient value of each event node in the drift analysis path within the time window, calculate the change rate of the stress gradient value within the time window, and get the vibration drift slope Sd;

[0132] Get the stress release drift slope Sr within the detection time window after the package is in the open and closed state;

[0133] Based on the material creep history data, set the drift threshold to k*Sd;

[0134] It needs to be explained that k is the multiple coefficient for amplifying the vibration drift slope (Sd);

[0135] If |Sr|>k*|Sd|, mark the current vibration sensor as a unpacking release type drift, otherwise mark it as a normal vibration drift;

[0136] If it is a unpacking release type drift, get the elastic modulus and viscous coefficient in the Burgers model parameters;

[0137] The elastic modulus and viscous coefficient are processed by ratio, and the time constant ti of creep hysteresis is obtained, which is used to analyze the hysteresis effect of stress release after unpacking;

[0138] It can be understood that analyzing the hysteresis effect of stress release after unpacking is beneficial to capturing the decay law of residual stress inside the goods over time, reducing the misjudgment of whether the goods have structural damage due to long-term creep only according to the stress measurement value at the unpacking moment; At the same time, the best time window for subsequent quality reinspection can be determined, and data support for optimizing the stress release performance of packaging materials can be provided, and the continuous impact of residual stress on goods during the hysteresis process is reduced;

[0139] Through the formula: Get the hysteresis impact factor Yz;

[0140] Where, Δt1 is the characteristic time constant of stress release;

[0141] It can be understood that when Δt1>3ti, the hysteresis effect is significant;

[0142] If the hysteresis effect is significant, set the creep sensitive area based on the event node distribution diagram, and obtain the creep risk nodes of the creep sensitive area in the drift analysis path;

[0143] Step four, based on the mutual correlation entropy of the creep risk node correlation entropy debarking stress correction model, combined with the mutual correlation entropy of the time window strong correlation node group appearance probability, the priority compensation time window is screened and the compensation strategy of the drift sensor is made;

[0144] Obtain the node feature vector of any two creep risk nodes, and calculate the mutual correlation entropy of the node feature vectors of the two creep risk nodes;

[0145] Wherein, the mutual correlation entropy is through the node feature vector of two creep risk nodes; Then through the kernel density estimation to fit the joint probability distribution of two vectors and their respective marginal probability distribution, and finally substitute into the mutual information formula to obtain:

[0146] Through the equation: The entropy value model is extended and corrected to obtain the corrected mutual correlation entropy H xy ;

[0147] It can be understood that the purpose of extending and correcting the entropy value model to obtain the corrected mutual correlation entropy is:

[0148] Purpose one is to reduce the interference factors not considered in the original entropy value model, such as false correlation caused by sensor drift and environmental noise, so that the corrected mutual correlation entropy reflects the real correlation strength between the creep risk nodes, and reduces the false judgment of strong correlation node groups due to model deviation;

[0149] Purpose two is to facilitate the dynamic adaptation of mutual correlation entropy to supply chain working condition changes, and to incorporate the time-varying law of node coordination characteristics through extension and correction, to provide more actual correlation quantization basis for subsequent compensation scheme based on strong correlation node groups, and to improve the adaptability of the compensation strategy;

[0150] Wherein, H is the mutual correlation entropy before correction, γ is the cargo sensitivity coefficient, Δδr and Δδt are the stress release amplitude and material yield strength threshold before and after debarking, respectively;

[0151] Based on the mutual correlation entropy H xy of any two creep risk nodes, the entropy value constraint modeling of the coordination compensation range is carried out, and the strong correlation node group sensitive to debarking is obtained;

[0152] Obtain the probability of occurrence of different strong correlation node groups in each time window in the supply chain process, and carry out information entropy value analysis on the time window to obtain the time node combination entropy;

[0153] Based on the time node combination entropy, a priority compensation time window is determined, and a measurement compensation scheme of the drift sensor is determined in the priority compensation time window;

[0154] As can be understood by those skilled in the art, by calculating the cross-correlation entropy of any two creep risk nodes, setting a correlation strength threshold for entropy value constraint modeling, defining the effective range of cooperative compensation, and screening out strong correlation node groups sensitive to the unpacking process;

[0155] The probabilities of each strong correlation node group appearing in each time window of the supply chain are counted, and the uncertainty of these probability distributions is calculated by using an information entropy algorithm to obtain time node combination entropy reflecting the time aggregation law of the node group;

[0156] According to the combination entropy value ranking, the lower the entropy value, the more stable the node group appears in the time window, and the time window with the maximum combination entropy value is determined as the priority compensation time window;

[0157] In the time window, a compensation scheme is developed for the drift sensor, such as a stress or displacement sensor with measurement deviation exceeding the standard, in combination with the cooperative characteristics of the strong correlation node group;

[0158] For example, if the strong correlation node group includes three sensors for monitoring the vibration of the goods in transport (high cross-correlation entropy, belonging to the unpacking sensitive group), the cooperative characteristics are as follows: under normal working conditions, the peak value deviation of the vibration acceleration of the three sensors is not more than 5%, and the fluctuation period is synchronized (because they are affected by the jolting of the transport vehicle);

[0159] When sensor B drifts, for example, the measured value is 15% higher than the actual value, a compensation scheme is developed based on the cooperative characteristics: taking the real-time average values of sensors A and C as the reference, and combining the peak value deviation coefficient Pz of B and A / C in the historical data;

[0160] The correction value B of B is calculated by the formula: new ;

[0161] Wherein, is the drift trend compensation term, which is determined based on the product of the historical drift slope of B and time, and SA and SC are the real-time average values of sensors A and C;

[0162] The peak value deviation coefficient is calculated by the ratio of the deviation amount of the average values of A and C to the average values in the historical data;

[0163] Meanwhile, the drift trend characteristics of B are retained in the compensated data, so that the data consistency of the node group can be checked during subsequent unpacking.

[0164] Embodiment 3

[0165] For example, Figure 3 ​As shown, the application is a supply chain-based sensor automatic detection system, comprising the following modules:

[0166] Hierarchical analysis module: used for obtaining transportation vibration energy of supply chain edge vibration sensors, performing gradient analysis on the transportation vibration energy to obtain hierarchical pressure gradient, establishing node fingerprint vector and event node distribution map based on the hierarchical pressure gradient;

[0167] Topology construction module: through sensing package opening and closing signals and strain data, creep coefficient reset trigger analysis is performed, if the reset is triggered, the event node distribution map is mapped into a weighted directed graph, the node transmission factor and path score of the weighted directed graph are extracted in combination with the material creep cumulative effect, and a risk potential model is constructed to output a path conduction topology graph;

[0168] Hysteresis positioning module: based on the path conduction topology graph, a drift analysis path is extracted, which is used to calculate the vibration drift slope and the stress release drift slope after unpacking in the drift analysis path, and to locate the creep sensitive area risk node with significant hysteresis effect;

[0169] Drift compensation module: based on the unpacking stress correction model of the creep risk node cross-correlation entropy, used for screening the priority compensation time window in combination with the cross-correlation entropy of the appearance probability of the strongly correlated node group in the time window, and formulating a compensation strategy for the drift sensor.

[0170] The above describes one embodiment of the application in detail, but the content described is only the preferred embodiment of the application, and cannot be considered as limiting the scope of the application. Any equivalent changes and improvements made in the scope of the application should still belong to the scope of the application.

Claims

1. A supply chain based sensor auto-detection method, characterized in that: The method comprises the following steps: obtaining transportation vibration energy of a supply chain edge vibration sensor, performing gradient analysis on the transportation vibration energy to obtain a hierarchical pressure gradient, establishing a node fingerprint vector and an event node distribution map based on the hierarchical pressure gradient; by sensing the package opening and closing signal and the strain data, performing creep coefficient reset trigger analysis, if the trigger reset is triggered, mapping the event node distribution map into a weighted directed graph, combining the material creep cumulative effect to extract the node transmission factor and the path score of the weighted directed graph, and constructing a risk potential model to output a path conduction topology graph; based on the path conduction topology graph, extracting a drift analysis path, calculating the vibration drift slope in the drift analysis path and the stress release drift slope after unpacking, and locating the creep sensitive area risk node with significant hysteresis effect; based on the unpacking stress correction model of the creep risk node mutual correlation entropy, combining the mutual correlation entropy of the appearance probability of the time window strong correlation node group to screen the priority compensation time window and formulate the compensation strategy of the drift sensor.

2. The method of claim 1, wherein: The manner of performing gradient analysis on the transportation vibration energy is: obtaining a three-dimensional pressure distribution field matrix of a cargo sensitive pressure load, extracting a gradient vector of the pressure distribution field by a central difference method, and constructing a hierarchical pressure gradient field; setting different feature sampling points for the hierarchical pressure gradient field, extracting texture features and level features of the sampling points, and performing event screening processing on the texture features and level features of the sampling points according to an event principle; marking the sampling points meeting the event principle as event nodes, performing space-time coding processing on the texture features and level features of the event nodes, and constructing a node fingerprint vector; obtaining all node fingerprint vectors of the cargo at different stages of the supply chain, and establishing an event node distribution map according to a node rule.

3. The method of claim 2, wherein: The manner of obtaining the pressure load of the cargo sensitive level is: obtaining vibration acceleration collected by a supply chain edge vibration sensor, performing multi-layer decomposition on the vibration acceleration signal by a discrete wavelet transform, and extracting an energy ratio of each layer to obtain a hierarchical energy ratio; constructing a stacked energy conversion model to convert the hierarchical energy ratio into pressure loads of different energy levels of a packaging contact surface during cargo transportation of the supply chain; screening the pressure loads of different levels to screen the pressure loads of the cargo sensitive level.

4. The method of claim 1, wherein: The manner of performing the creep coefficient reset trigger analysis is: obtaining a package opening and closing signal, and determining an opening and closing state of the package based on the package opening and closing signal; if the package is in the opening and closing state, triggering a creep coefficient reset algorithm to reset the creep coefficient to obtain a real-time creep coefficient.

5. The method of claim 1, wherein: The manner of constructing the risk potential model to output the path conduction topology graph is: obtaining a weighted directed graph and a node fingerprint vector of the weighted directed graph, combining the material creep cumulative effect to extract the node transmission factor and the path score; constructing a risk potential model based on the node transmission factor and the path score, extracting a conduction efficiency and a key path of creep risk through the risk potential model; based on the obtained key path, outputting a path conduction topology graph to an upstream and downstream of the supply chain.

6. The method of claim 5, wherein: The manner of obtaining the weighted directed graph is: obtaining a package state vector constructed by a real-time creep coefficient, a stress release coefficient and a package opening and closing state; Obtaining a timestamp of the package state vector and a timestamp of the node fingerprint vector, aligning and splicing the package state vector and the node fingerprint vector based on the timestamps, and establishing a node feature vector; Collecting the weight between any two event nodes in the event node distribution diagram, and mapping the event node distribution diagram into a weighted directed graph.

7. The method of claim 6, wherein: The extraction method of the weight between the two nodes is: Obtaining any two event nodes in the event node distribution diagram, if the two event nodes are in the same isochronous node group, calculating the Euclidean distance of the node feature vectors of the two event nodes as the node judgment distance; If the two event nodes are not in the same isochronous node group, or the node judgment distance is higher than or equal to the preset node judgment distance threshold, directly performing node stress conduction analysis to obtain the pressure gradient difference; Calculating the spatio-temporal correlation degree of the node feature vectors and the timestamps of any two nodes; Building a weight equation, inputting the pressure gradient difference, the spatio-temporal correlation degree and the real-time creep coefficient into the weight equation to obtain the weight between any two nodes.

8. The method of claim 1, wherein: The way to locate the creep sensitive area risk node with significant hysteresis effect is: Extracting the path with the maximum potential energy accumulation and the continuously increasing creep coefficient from the path conduction topology diagram as the drift analysis path; Obtaining the pressure gradient value of each event node in the drift analysis path within the time window, calculating the change rate of the pressure gradient value within the time window to obtain the vibration drift slope; Obtaining the stress release drift slope within the detection time window after the package is in the open and closed state; Determining the drift type based on the vibration drift slope and the release drift slope, if it is a package release type drift, obtaining the hysteresis influence factor and judging the significant hysteresis effect; If the hysteresis effect is significant, setting the creep sensitive area based on the event node distribution diagram, and the creep risk node in the creep sensitive area in the drift analysis path.

9. The supply chain based sensor automated detection system of claim 1, wherein: The way to screen the priority compensation time window is: Obtaining the node feature vectors of any two creep risk nodes, calculating the cross-correlation entropy of the node feature vectors of the two creep risk nodes, and extending and correcting the cross-correlation entropy to obtain the corrected cross-correlation entropy; Based on the cross-correlation entropy of any two creep risk nodes, entropy value constraint modeling of the cooperative compensation range is carried out to obtain a strong correlation node group sensitive to package opening; Obtaining the probability of the appearance of different strong correlation node groups in each time window in the supply chain process, and performing information entropy value analysis on the time window to obtain the time node combination entropy, Based on the time node combination entropy, the priority compensation time window is determined.

10. The supply chain based sensor automated detection system of claim 1, wherein: Comprise the following modules: The hierarchical analysis module is used to obtain the transportation vibration energy of the supply chain edge vibration sensor, perform gradient analysis on the transportation vibration energy to obtain hierarchical pressure gradient, and establish node fingerprint vector and event node distribution diagram based on the hierarchical pressure gradient; The topology construction module extracts the node transmission factor and path score of the weighted directed graph by sensing the package opening signal and strain data, and constructs a risk potential model to output a path conduction topology diagram. Hysteresis positioning module: based on the path conduction topology graph to extract the drift analysis path, used to calculate the vibration drift slope in the drift analysis path and the release drift slope after unpacking the stress, and locate the creep sensitive area risk node with significant hysteresis effect; Drift compensation module: based on the unpacking stress correction model of the creep risk node mutual correlation entropy, used to combine the mutual correlation entropy of the appearance probability of the strong correlation node group in the time window, screen the priority compensation time window and develop the compensation strategy of the drift sensor.

Images