A method for tracing the authenticity of mechanical tests

Abstract

The present invention relates to the field of engineering inspection information technology, and in particular to a method for tracing the authenticity of mechanical tests. The method comprises the following steps: a sample is dual-anchored by nucleic acid barcode-three-dimensional scanning to generate a unique identifier and encapsulate a first data packet; an event stream is mapped into a third-order sparse tensor, a topological barcode is extracted, and a second data packet is generated using a threshold signature; a Graphene ordinary differential equation outputs a consistency score, and a gain ring structure is used to generate a third data packet in the event of an anomaly; the core information of the three data packets is aggregated and written into a blockchain via Groth16 zero-knowledge proof, and an audit node automatically verifies and issues an electronic certificate of conformity or records violations, ultimately forming a queryable traceability report to achieve full-process anti-counterfeiting, real-time early warning, and publicly verifiable auditing.

Description

Technical Field

[0001] The present invention relates to the field of engineering detection information technology, and in particular to a mechanical test authenticity tracing method. Background Art

[0002] In transportation engineering, the results of mechanical tests on large-volume concrete, reinforced components, and asphalt mixtures directly determine the structural safety and lifespan of concealed projects. If test samples are replaced, data is tampered with, or reports are falsified, unqualified materials will be mixed into the project, causing serious quality accidents and increasing subsequent maintenance costs. The industry has adopted QR codes, radio frequency tags, database logs, and other means for traceability, but these methods only cover a single link: QR codes are easily copied and replaced, and radio frequency tags become ineffective in high temperature or humid environments; centralized databases lack cross-institutional consensus, and logs can be rewritten by administrators. When there is room for cross-cheating between "people," "machines," "materials," "methods," and "loops" in a multi-factor chain, existing technologies cannot verify whether the data has been tampered with within the process, nor can they allow third parties to quickly review without disclosing commercial secrets. Summary of the Invention

[0003] In response to the many problems existing in the above-mentioned existing technologies, the present invention provides a method for tracing the authenticity of mechanical tests. The present invention generates a unique identifier by dual anchoring of nucleic acid barcode and three-dimensional model; the real-time event flow through the time window is mapped into a third-order sparse tensor and the topological barcode is extracted; the Graph God ordinary differential equation is combined with the topological prior to output the consistency score and dynamically gain the abnormal loop; the three-stage data is aggregated and written into the chain through threshold signature and Groth16 proof to form a publicly verifiable electronic certificate, thereby achieving the effects of full-process anti-counterfeiting, real-time warning and automatic audit.

[0004] A method for tracing the authenticity of a mechanical test comprises: applying a unique identifier to a specimen and synchronously acquiring three-dimensional measurement data to generate fingerprint data and a three-dimensional model; hashing the two to obtain a unique identifier and erasure coding; collecting personnel, equipment, process, environment, and sample status information under a unified time base to form an event stream, which is encapsulated into a first data packet; mapping the event stream into a third-order sparse tensor within a set window, extracting a topological barcode vector, generating a consensus signature using a threshold signature, and encapsulating the signature together with the unique identifier into a second data packet; solving a graph-based ordinary differential equation containing topological constraints based on the third-order sparse tensor and the event stream to obtain a consistency score; if the consistency score is lower than a threshold, adjusting the topological barcode vector and hyperedge weight to generate an adjustment log, and encapsulating the signature together with the consistency score into a third data packet; using the third-order sparse tensor, topological barcode vector, consensus signature, consistency score, and adjustment log as private inputs, generating and aggregating a zero-knowledge proof and submitting it to a distributed ledger; writing an electronic certificate of conformity or recording violation information after verification by an audit node; integrating the first data packet, the second data packet, the third data packet, the third data packet, and the aggregated proof to generate a traceability report and provide a query.

[0005] Preferably, when applying a unique identification to the sample, nucleic acid barcode particles are deposited by spraying, and the nucleic acid barcode particles produce a unique fluorescence response in the ultraviolet band; the sample fingerprint data and the sample three-dimensional model data are recorded with the same timestamp and spatial coordinates before packaging.

[0006] Preferably, when mapping the event stream into a third-order sparse tensor, a fixed-length time window is used to sort the events, and three event records with a time difference not greater than a preset threshold and a consistent source are written into the third-order sparse tensor as a tensor index.

[0007] Preferably, when extracting the topological barcode vector, a Vietoris–Rips complex is constructed for the third-order sparse tensor, and zero-dimensional, one-dimensional and two-dimensional topological invariants are calculated and combined in order of appearance to form the topological barcode vector.

[0008] Preferably, a threshold signature algorithm is applied when generating a consensus signature, and the threshold value is equal to the total number of witness nodes minus the number of tolerable failure nodes plus one.

[0009] Preferably, before solving the graph neural ordinary differential equation, a node feature vector is constructed for each event based on the third-order sparse tensor and the event stream data, and the time difference feature is standardized to form a node feature matrix; the graph convolution kernel adopts the GraphSAGE structure.

[0010] Preferably, when the consistency score is lower than the threshold, the element representing the one-dimensional ring structure in the topological barcode vector is multiplied by a gain coefficient, and the hyperedge weight associated with the element is adjusted by the same coefficient to generate an adjustment log.

[0011] Preferably, the Groth16 proof scheme is adopted when generating the zero-knowledge proof, and the third-order sparse tensor, topological barcode vector, consensus signature, consistency score and adjustment log are used as private inputs, and the hash value of the above private inputs is used as public input.

[0012] Preferably, multiple zero-knowledge proofs are aggregated into a single aggregate proof through elliptic curve homomorphism, and a signature is generated by a trusted execution environment before being submitted to the distributed ledger.

[0013] Preferably, the timestamp of the event stream is generated by a time synchronization system compliant with the IEEE1588 protocol and written along with the event.

[0014] Compared with the prior art, the advantages and beneficial effects of the present invention are:

[0015] Through dual-source unique identification generation and erasure coding technology, the inseparable binding of the sample's microscopic fingerprint and three-dimensional morphology is achieved.

[0016] Through the third-order sparse tensor-topological barcode mapping and threshold signature technology, cross-node event structure consensus and high-reliability tampering detection are achieved.

[0017] Through the self-supervised evaluation technology of graph-based ordinary differential equations with topological constraints, real-time quantification and adaptive amplification of hidden anomalies in the experimental process are achieved.

[0018] Through the Groth16 zero-knowledge proof aggregation and chain technology, leak-free public verifiable auditing and automatic issuance of electronic certificates are achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 Schematic diagram of the process of the present invention;

[0020] Figure 2 Schematic diagram of zero-knowledge proof aggregation and ledger interaction in the method of the present invention. DETAILED DESCRIPTION

[0021] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0022] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise," "include," etc. used herein indicate the presence of the features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0023] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification.

[0024] like Figure 1 As shown, a mechanical test authenticity tracing method includes:

[0025] A unique identifier is applied to the specimen, and three-dimensional measurement data is simultaneously acquired to generate fingerprint data and a three-dimensional model. The two are hashed to obtain a unique identifier and erasure-corrected encoding. Information on personnel, equipment, process, environment, and sample status is collected under a unified time base to form an event stream, which is encapsulated as a first data packet. The first step of this invention focuses on uniquely anchoring the specimen before mechanical testing and closing the data loop for the entire process. Its core principle is to utilize the "dual-source fusion" principle of microscopic molecular identifiers and macroscopic geometric information to transform the physical object into an unalterable digital entity and embed it into the subsequent traceability chain. During implementation, prefabricated nucleic acid barcode particles are first deposited onto the specimen surface using microspray technology. These nucleic acid barcode particles are composed of oligonucleotide sequences of fixed length, and different sequences exhibit unique fluorescence peak ratio characteristics under ultraviolet narrowband excitation. Spectral acquisition and convolutional neural network decoding yield a 128-bit sequence, referred to herein as fingerprint data. Because nucleic acid sequences are random and difficult to physically replicate, fingerprint data can be considered an unclonable signature at the microscopic level.

[0026] Subsequently, a laser 3D scanner is used to perform full-field measurement on the same sample. The scanned point cloud is subjected to posture registration, noise filtering and resampling processing to obtain a 3D model facing the loading direction. This 3D model reflects the macroscopic morphology, and its detail resolution is better than 0.05 mm. It can form a stable geometric texture mapping on the surface of materials such as concrete and steel bars. The fingerprint data and the 3D model data are written with a unified timestamp and position coordinates under the same clock signal to achieve acquisition synchronization. In order to further ensure the consistency of subsequent verification, the present invention performs hash operations on the two sets of data in a trusted execution environment. The core calculation expression is: ,in Indicates a unique identifier. A binary string representing the fingerprint data, A binary string representing 3D model data, " is the concatenation operator. This expression uses the one-way and avalanche properties of the secure hash function to make any pair or Small changes will result in The overall changes provide accurate and highly sensitive comparison basis for subsequent on-chain verification.

[0027] The unique identifier generated enters the redundant storage process. The present invention uses Reed-Solomon erasure coding to The code is encoded into three erasure-correction fragments. This algorithm allows any two of the k fragments to reconstruct the original code, enabling complete recovery of unique identifiers even in the event of node offline or network partition. The three fragments are stored in a local laboratory node, a node on the project owner's side, and an offline node on the review side, eliminating single points of failure.

[0028] To ensure causal consistency throughout the entire process, the present invention utilizes a hardware time synchronization system compliant with the IEEE 1588 protocol. Synchronous pulses are broadcast at a one-second interval to the nucleic acid calibration equipment, 3D scanning equipment, and status acquisition system. All data is written with a pulse number and an internal micro-time sequence number. This allows fingerprint data to be bound to the 3D model data. Subsequently, collected data on personnel identity, equipment operation, process settings, environmental monitoring, and real-time sample status are also arranged in a unified time sequence, forming an event stream. This event stream is stored in a circular buffer with a write rate of 10 Hz, ensuring stable sequencing despite network jitter.

[0029] Finally, the fingerprint data, 3D model data, unique identifier, erasure fragments, and event stream are encapsulated into the first data packet according to the predefined packet structure. The packet header contains the protocol version and total length fields; the index area provides the unique identifier and fragment location index; the payload area contains the raw measurement data in field order; and the trailer contains an elliptic curve digital signature generated by the trusted execution environment. The digital signature is combined with the packet body hash and timestamp to provide tamper detection capabilities.

[0030] In a typical implementation, the nucleic acid barcode spraying and scanning process for a concrete cubic test block for a highway bridge took a combined 25 seconds; unique identifier calculation and erasure coding took a combined 5 milliseconds; event stream collection was completed 30 seconds before the loader started, generating a first data packet of 2 megabytes. The package signature was written in a local private chain channel, and subsequent steps used the package hash as the parent node for topological mapping and zero-knowledge proof.

[0031] In terms of effect, dual-source hashing greatly improves the robustness of sample identity verification; erasure storage reduces the probability of identity loss due to network disconnection; global time synchronization provides a strict causal order for the subsequent third-order sparse tensor construction. In addition, the combination of digital signatures and salt value randomization can resist conventional playback and replay attacks. Comprehensive experimental results show that compared with the single-source solution that only uses QR code or radio frequency identification technology, the present invention reduces the false detection rate by 60% and the missed detection rate by 80% in the sample replacement detection scenario. It is a high-reliability digital traceability method suitable for the full-cycle review of traffic engineering mechanics tests.

[0032] Preferably, when applying a unique identification to the sample, nucleic acid barcode particles are deposited by spraying, and the nucleic acid barcode particles produce a unique fluorescence response in the ultraviolet band; the sample fingerprint data and the sample three-dimensional model data are recorded with the same timestamp and spatial coordinates before packaging.

[0033] The main purpose of using a spray method to deposit nucleic acid barcode particles is to construct random and non-replicable microscopic sequence identifiers on the surface of the sample. Each particle is composed of prefabricated oligonucleotides, and their base arrangement is randomly extracted through an in-library algorithm to ensure that the sequence within the same batch is not repeated. When the particles are excited by ultraviolet light with a wavelength of 365nm, fluorescence peaks will be generated at both 580nm and 680nm. The ratio of the two peak intensities corresponds to a unique sequence key. A two-dimensional spectral matrix can be obtained by synchronous scanning using a narrow-band excitation light source and a hyperspectral camera. The convolutional neural network extracts the peak ratio of the matrix pixel by pixel, and finally obtains a digital sequence of 128 bits in length, which is referred to as fingerprint data in this article. The oligonucleotide chain requires specific enzyme cutting and temperature control conditions to break. The sequence cannot be destroyed by daily handling, so the fingerprint data is inherently non-cloneable.

[0034] Macro-geometric information is acquired through a single laser scan. The scanner samples the entire specimen surface with a 950nm pulse, with an average point spacing of 0.05mm, generating a dense point cloud. After iterative closest point registration and outlier filtering, the point cloud generates a 3D model oriented in the loading direction and saved as vertex-normal pairs. The scanner's built-in encoder automatically converts the original polar coordinates to the device's Cartesian coordinate system, ensuring consistent spatial reference between measurements.

[0035] To bind the microscopic sequence with the macroscopic morphology, the present invention shares the same pulse clock signal in the two data acquisition links. The same timestamp and scanning reference coordinates are written into the fingerprint data file header and the 3D model file header. The secure hash function is then called to perform one-way compression on the two sets of data. The core calculation formula is: ,in Indicates a unique identifier. Represents the fingerprint data binary string, Represents a three-dimensional model binary string, " is the concatenation character. The secure hash function has an avalanche effect. or Small changes will result in Overall changes, therefore Inheriting both microscopic and macroscopic information. To improve storage toughness, the present invention will Input Reed-Solomon encoder to generate three erasure fragments , any two pieces can be restored .

[0036] Next, five types of status information are collected at a 10Hz frequency using a unified time base: personnel identity using a universal FIDO2 key, equipment operation records using a boot hash, process parameters recording loading speed and target stress level, environmental monitoring recording temperature and humidity, and sample status recording fluorescence decay curves. Each record is timestamped and identified by the source node and written sequentially to the event stream buffer.

[0037] Finally, the fingerprint data, 3D model data, unique identifier, erasure fragments and event stream are encapsulated together into the first data packet. The packet structure is divided into four areas: the packet header contains the protocol version and packet length, the index area stores The payload area contains raw measurement data and event streams, along with fragment offsets. The end contains an elliptic curve digital signature generated by the trusted execution environment. The digital signature encrypts the packet body hash, and any node can verify its integrity using the public key.

[0038] Example: Spraying a nucleic acid barcode on a cubic bridge concrete test block takes 15 seconds, spectrum acquisition and convolution decoding takes 5 seconds, laser scanning and registration takes 20 seconds, hashing and erasure coding takes 10ms, and the total packaging process latency is less than 50 seconds. Before the loader is started, the review node verifies the signature using the public key in the package, and there is no sign of tampering. Re-verification is performed after 3 months, and the barcode is rescanned and the hash value is recalculated. .like Determine that the sample has not been replaced; if Automatically alert the police and locate the person responsible.

[0039] Compared to QR codes or RFID tags, this solution requires no external power supply, is difficult to copy, and is bound to topography and texture. Dual-source hashing and fragment storage prevent identity loss. A unified clock assigns causal order to data, laying the foundation for subsequent third-order sparse tensor and topological analysis. Random inspections of 100 batches of specimens showed an increase in the replacement detection hit rate to 99%, with a false alarm rate of less than 1%, significantly improving the reliability of authenticity traceability in traffic engineering mechanics testing.

[0040] Map the event stream into a third-order sparse tensor within the set window, extract the topological barcode vector, generate a consensus signature using the threshold signature, and encapsulate it into a second data packet with a unique identifier; in the core link of the present invention, the system first sets a time window of fixed length and writes the event stream obtained in the previous step into the buffer in the order of occurrence. Each event record contains a source identifier, a timestamp, and a five-element state vector. In order to capture the correlation between the timing and source of the three events, the present invention introduces a third-order sparse tensor structure. Given the event sequence within the window If the following conditions are met: 1. The time difference between two adjacent events is not greater than the preset threshold; 2. The source identifiers of the three events are the same, then an index triplet is constructed. , in tensor coordinates Disposal , the remaining elements are The present invention adopts coordinate format to store To save memory. Represents a third-order sparse tensor, letter Indicates the sequence number of the event within the window.

[0041] After the sparse tensor is constructed, the topological homology method is used to extract high-dimensional geometric information. Create a Vietoris–Rips complex, set the growth radius step, and calculate the zero-dimensional, one-dimensional, and two-dimensional topological invariants. The output is organized into vectors in the order of dimensions. , called a topological barcode vector. The topological barcode vector can reveal the relationship between the connectivity, ring structure and cavity between events, and provide a structural prior for subsequent graph neural reasoning. In order to prevent the sparse tensor and topological barcode from being tampered with during network transmission, this paper introduces a threshold signature algorithm to generate consensus signature data. Assume that the total number of witness nodes is , the number of tolerable failed nodes is , the threshold value is Each node pair Calculate partial signature When a partial signature of at least the threshold value is collected, a consensus signature is obtained using a public aggregation function. The simplified expression is: , where the letters Indicates consensus signature data, letters Indicates the The threshold signature mechanism ensures that the leakage of a single node's private key will not lead to the forgery of the overall signature. Only when the threshold number of witness nodes jointly sign can a verifiable signature be obtained. .

[0042] Then, the system converts the third-order sparse tensor , topological barcode vector , consensus signature data With unique identifier Perform index coupling to generate the second data packet. The packet structure consists of a header, an index area, a payload area, and a digital signature area. The index area stores and The hash value of the payload area is written and The digital signature area uses the reviewer's key to sign the package body hash for on-chain verification.

[0043] In terms of effect, the third-order sparse tensor captures the three-event correlation within the window, and can express the multi-dimensional coupling of "man-machine-method" in the loading process; the topological barcode vector compresses the high-dimensional connection relationship into a low-dimensional vector, which is quickly read by the subsequent ordinary differential graph network engine; the threshold signature ensures data integrity and non-repudiation; the second data packet is bound to a unique identifier, so that the physical sample, original measurement data and high-dimensional topological features form a closed chain.

[0044] Example: Assume the window length is 200ms and the number of events ; The tensor construction yields approximately 1500 non-zero indices, and the storage occupies 0.3% of the dense tensor volume; the zero-dimensional, one-dimensional, and two-dimensional topological invariants are 5, 12, and 2, respectively, corresponding to ,exist Collect 5 partial signatures on the threshold network and aggregate them to generate The second data packet size is 320kB. After the review node verifies that the signature is valid, the chain is written. If the tensor or topology barcode is replaced, Failure to pass public key verification directly triggers a traceability exception. In actual testing, 1,000 random tampering attempts were verified, with a false positive rate of 0 and a missed positive rate of less than 1%.

[0045] Preferably, when mapping the event stream into a third-order sparse tensor, a fixed-length time window is used to sort the events, and three event records with a time difference not greater than a preset threshold and a consistent source are written into the third-order sparse tensor as a tensor index.

[0046] After completing the first data packet encapsulation, the system enters the event structure encoding phase. The goal of this phase is to transform the continuously generated event stream into a third-order sparse tensor that reflects the timing-source coupling relationship and extract the topological barcode vector corresponding to the tensor topology to provide a structural prior for subsequent dynamic consistency scoring.

[0047] The event stream consists of multi-dimensional state records, each of which contains a source identifier, a UTC timestamp, and a five-element state vector. The system first sets a fixed-length time window, such as 200ms, and numbers the events that fall into the same window in the order of arrival. To avoid duplicate writes due to window overlap, the window is scrolled with a sliding step equal to the window length.

[0048] The mapping rules are based on the two criteria of "time difference" and "source consistency". They belong to the same window and have the same source identifier, and satisfy the following formula: , then in the third-order sparse tensor Index The value of 1 is assigned to each element, and the other elements remain 0. is the event timestamp, is the threshold constant. Since the number of events is much larger than the number of trigger condition combinations, There are very few non-zero coordinates, and the coordinate format only needs to record the triplet With a value of 1, the memory usage can be controlled to one thousandth of the fully expanded volume of the window event.

[0049] After completing the tensor construction, the system calls the topological homology library on the GPU to Construct the Vietoris–Rips complex. The complex uses the relative sequence difference between events as the pseudo-distance, and extracts 0-dimensional connected branches, 1-dimensional ring structures, and 2-dimensional cavities by gradually increasing the radius. The stable quantities that remain unchanged in the continuous radius interval of each dimension are combined in the order of dimensions to obtain the topological barcode vector For example, in a bridge compressive loading test, a window was , indicating that there are four connected components, 11 independent loops, and one two-dimensional hole within this interval. Loops and holes typically correspond to cyclic dependencies between personnel and equipment operations or data gaps. This invention will utilize these features to identify violations in the next stage.

[0050] To ensure and The present invention applies the threshold signature algorithm to generate consensus signature data without being tampered with during link transmission or local cache. Assuming the total number of witness nodes is 7 and the number of offline nodes can be 2, the threshold is 5. Each witness node uses an independent private key to log messages. Calculate partial signature After collecting 5 partial signatures, perform an aggregation operation: , where the letters Indicates consensus signature data, letters Indicates the Partial signatures of witness nodes. Aggregation can be completed with only one broadcast-aggregation-verification process without introducing a lot of interaction delays. Any node with less than the threshold number cannot forge a valid .

[0051] The system then writes the third-order sparse tensor, topological barcode vector, consensus signature data and unique identifier into the second data packet in the preset field order, and the end of the packet is signed by the review agency's elliptic curve private key to form a double integrity protection. After receiving the second data packet, other nodes on the chain first verify the outer signature with the review public key, and then use the witness network public key set to verify , if both steps pass, the tensor-topology pair is accepted.

[0052] In terms of technical effects, the third-order sparse tensor explicitly encodes the combination relationship of three events, which not only retains the micro-loop information during the loading process, but also avoids the storage explosion caused by high-dimensional full connection; the topological barcode vector strongly compresses the tensor, which is convenient for the subsequent ordinary differential graph model input; the threshold signature provides anti-repudiation and anti-single-point forgery capabilities; the second data packet is coupled with the unique identifier to ensure that the sample identity, original data and structural characteristics are inseparable.

[0053] In a comparative test of 100 on-site stress tests, the proposed solution achieved an average detection accuracy of 0.97 for human frame skipping and false device number insertion, while the traditional time series isolation forest method was only 0.72. At the same time, the processing time for each window did not exceed 20ms, meeting real-time monitoring needs.

[0054] Preferably, when extracting the topological barcode vector, a Vietoris–Rips complex is constructed for the third-order sparse tensor, and zero-dimensional, one-dimensional and two-dimensional topological invariants are calculated and combined in order of appearance to form the topological barcode vector.

[0055] In the context of mechanical test authenticity tracing, a third-order sparse tensor preserves the temporal order and source relationships of three event groups within a window period. However, the tensor itself has high dimensionality and a low nonzero ratio. Directly using it for subsequent reasoning affects computational efficiency and makes it difficult to intuitively reveal potential illegal structures. This paper introduces topological coherence to compress the connectivity, loop, and cavity features in the tensor into short vectors, thereby preserving structural information while reducing feature dimensionality.

[0056] First, we give the core mathematical expression. Let the sparse tensor be , construct the Vietoris–Rips complex. The complex has the radius parameter The vertex set under is the event index, and the edge set is the two points with a "pseudo distance" less than Determine, triangles and higher dimensional simplexes are generated recursively. Increase, the complex topology changes, the number of zero-dimensional connected branches , one-dimensional ring number , the number of two-dimensional cavities The present invention records the first stable value that appears in the radius sequence and lasts for more than three steps, and combines them into a topological barcode vector: ,in 、 、 Represent the number of connected branches, ring structures, and cavities, respectively. Since common violations in mechanical testing (such as personnel replacement, equipment disconnection, and step skipping) will change the "pairing" relationship between events, and thus change the number of rings or cavities, the vector It can be considered as a sensitive fingerprint of abnormal patterns.

[0057] At the implementation level, to avoid high-dimensional simplex explosion, the present invention defines "pseudo-distance" as the maximum value of event index difference and limits the radius step to only 3 discrete values. Therefore, the scale of the complex is linearly related to the number of window events, and even in high-frequency loading experiments with dense events, the coherence calculation can be completed in real time on the GPU. The sequence is stored in the buffer, and the subsequent ordinary differentiable graph network is directly read by thread index without additional format conversion.

[0058] For example, during the high-pressure phase of a bridge concrete loading experiment, 120 events were generated within a 200ms window. After constructing the complex, the stable value Among them, 11 ring structures correspond to the cyclic dependence between the equipment status and the ambient temperature and humidity in a short period of time, while the single two-dimensional cavity reflects a blank area where both the equipment and personnel status are missing. The subsequent consistency scoring model will The peak and The occurrence of is determined as a high-risk window and triggers the topology gain mechanism to increase the sensitivity to illegal signals. The comparative experiment shows that when a fake device number is inserted, From 11 to 5, The model consistency score dropped significantly, and the abnormal data packet was successfully intercepted.

[0059] In terms of effect evaluation, after using topological barcode vector, the tensor feature dimension is theoretically It is reduced to 3, which reduces the input bandwidth of the subsequent graph god ordinary differential equations; at the same time, because rings and cavities are extremely sensitive to data gaps, in hundreds of sets of random tampering tests, the average detection rate of this method for two types of covert operations, "silent insertion" and "underreported events", reached 0.97, while the traditional time series statistics based on global mean or peak count was only 0.71.

[0060] Preferably, a threshold signature algorithm is applied when generating a consensus signature, and the threshold value is equal to the total number of witness nodes minus the number of tolerable failure nodes plus one.

[0061] In the second data packet generation phase, the present invention uses the threshold signature algorithm to perform consensus authentication on the third-order sparse tensor and topological barcode vector to resist the risk of tampering caused by single-point private key leakage and node offline. Threshold signature belongs to the category of distributed digital signature. Its basic idea is to split the elliptic curve private key originally stored in a single node into several private key fragments, which are respectively distributed to the nodes. Witness nodes; only when no less than the threshold value t valid partial signatures are collected, the aggregation function can output a complete signature that meets the public key verification rules. The selection of the threshold value is directly related to the network scale and the number of tolerable failure nodes. The present invention provides a determination formula: ,in Indicates the total number of witness nodes; Indicates the number of nodes that are allowed to be offline or compromised at the same time; Indicates the minimum number of partial signatures required to generate a valid consensus signature.

[0062] This formula ensures that the system can It can still work normally even if the number of nodes fails. At the same time, when the number of nodes controlled by the attacker does not exceed The signature cannot be forged at any time, thus achieving a provable balance between usability and security.

[0063] The generation process is as follows. First, in the key generation ceremony, the polynomial secret sharing algorithm is used to split the elliptic curve master private key into The private key fragments are written into the hardware security module of each witness node. and topological barcode vectors After that, the system will digest the message Broadcast to the witness network. Each online node calls the hardware security module to perform local exponential operation on the digest to obtain a partial signature. When the aggregation node collects no less than share When , an elliptic curve homomorphic summation is performed according to the Lagrange weight coefficient to obtain the complete signature The entire process only requires one round of message broadcasting and one aggregation operation, and the network complexity is low.

[0064] At the application level, in order to adapt to the traffic engineering test scenario, the present invention adds seven nodes, including the owner unit, supervision unit, construction laboratory, design unit, quality supervision platform and two redundant servers, to the witness network; the number of offline nodes tolerated is set to 2, and the threshold value The calculation is 6. This means that even if two nodes are temporarily unavailable due to network disconnection or maintenance, the remaining five nodes can still generate legitimate signatures; at the same time, if the attacker only controls two nodes, he cannot export , improve the anti-counterfeiting strength.

[0065] The technical effect is reflected in three aspects. First, anti-single point failure: the failure of a single node will not block the signature process, ensuring high availability. Second, anti-private key leakage: attackers need to break through at least one hardware security module at the same time to have the opportunity to forge a signature, greatly improving the security boundary. Third, low verification overhead: after aggregation, The size is the same as that of an ordinary single signature, and on-chain nodes only need one elliptic curve multiplication to verify it, without increasing bandwidth and computing power.

[0066] In a concrete cubic pressure test on a highway bridge, the total number of witness nodes , 2 offline nodes still meet the threshold After the system broadcasts the window summary, it collects 6 partial signatures in an average of 5 milliseconds, and the aggregation operation takes 7 milliseconds. The size is 96 bytes, and then the second data packet is written. The experimenter deliberately disconnected the two nodes from the network, and the next window still generated a new one within 12 milliseconds. If a tensor triple is randomly modified, the resulting signature cannot be verified by any node's public key due to a digest discrepancy, triggering a real-time traceback alert. After 1,000 random tampering tests, the system achieved a zero false positive rate and a low false negative rate of less than 1%.

[0067] Based on the third-order sparse tensor and event stream, a graph-based ordinary differential equation containing topological constraints is solved to obtain a consistency score. If the consistency score is lower than a threshold, the topological barcode vector and hyperedge weight are adjusted to generate an adjustment log, which is packaged into a third data packet with the consistency score. In the third stage, the present invention uses the graph-based ordinary differential equation to dynamically evaluate the consistency of window period event behavior, and modifies the topological barcode vector and hyperedge weight in real time after detecting an anomaly to generate a third data packet. In this stage, the structural information carried by the third-order sparse tensor is input in conjunction with the timing information of the original event stream to synchronously measure the degree of convergence between the "ought topology" and the "actual behavior."

[0068] The core of the graph ordinary differential equation is to regard the evolution of node state over time as a continuous trajectory. Let the number of events in the window be , the node feature matrix is ​​recorded as ,in Concatenate the source one-hot vector, the five-element state vector, and the time difference feature and normalize them. Hidden state matrix by As initial conditions, in the continuous time domain Satisfies the following differential equation: , where represents a third-order sparse tensor; is the topological barcode vector; For the three-layer GraphSAGE convolution kernel, the parameter set Obtained through prior training; and Used to inject topological external forces.

[0069] The system uses an explicit fifth-order Runge-Kutta integration method on the GPU, with a step size of . The integration ends and we get To measure the deviation between the measured behavior and the model expectation, the present invention designs a self-supervised loss: ,in is the Frobenius norm, is the balance coefficient, is the historical steady-state mean. The first term measures the reconstruction error between the node representation and the input after evolution; the second term measures the distance between the topological barcode vector and the long-term equilibrium state. The consistency score is obtained by mapping it to the interval (0,1) through the Sigmoid function In the offline stage, the threshold is selected according to the material type and loading method. When the window is running, if , indicating that the behavior has significantly deviated from the steady state.

[0070] After the deviation occurs, the present invention adopts the topological feedback mechanism to make the model online adaptive: first, the elements representing the one-dimensional ring structure in the topological barcode vector are Multiply by the gain factor ; Then find the tensor with For all corresponding hyperedges, multiply their weights by the same coefficient. The system writes the timestamp, gain index, coefficient, and pre- and post-adjustment values ​​to the adjustment log.

[0071] Consistency score Together with the adjustment log, it is encapsulated into the third data packet. The packet header contains the protocol version and length; the index area stores the unique identifier hash; the payload area is written The data is then stored in a log file and the elliptic curve digital signature of the trusted execution environment is appended to the end. Nodes on the chain verify the signature to confirm that the data has not been tampered with and can be traced back to the corresponding window.

[0072] In terms of technical effectiveness, continuous-time modeling avoids discrete step accumulation errors and can capture sub-second rate anomalies. Topological external forces make the model highly sensitive to changes in connectivity, rings, and cavities. The online gain mechanism maintains model robustness while significantly reducing the missed detection rate. Experimental statistics show that in a stress test involving the insertion of a forged person event, the system lowered the score to below 0.5 and logged it within an average of two windows (400ms), while it took six windows to reach the same threshold without topological feedback.

[0073] Example: The bridge shear-compression loading test window length is set to 200ms, and the total number of windows is 300. The average score of the normal window is 0.97, and the standard deviation is 0.03. When the loader number is replaced in the 150th window, the score drops to 0.60; the system will The gain was 1.2x, adjusting the weights of 32 hyperedges. The third data packet was 14kB in size and was written to the chain within 30ms. When the same anomaly reappeared, the score in the first window dropped to 0.48, triggering an alarm at the audit node. Compared to traditional mean filtering anomaly detection, the average response time was shortened by 68%, and the false alarm rate was reduced by 40%.

[0074] Preferably, before solving the graph neural ordinary differential equation, a node feature vector is constructed for each event based on the third-order sparse tensor and the event stream data, and the time difference feature is standardized to form a node feature matrix; the graph convolution kernel adopts the GraphSAGE structure.

[0075] This stage aims to transform the event stream into a node representation usable in a continuous-time graph model to capture the microscopic differences in mechanical testing operations. First, the pairing relationships of events within the window are obtained based on a third-order sparse tensor. This is used as an index to construct a node feature vector for each event. The node feature vector consists of three parts: a one-hot encoding of the source identifier, ensuring that events from the same source can "perceive" each other during convolutional aggregation.

[0076] The five-element state vector includes personnel identification code, equipment operation flag, process parameter summary, quantified values ​​of ambient temperature and humidity, and instantaneous sample strain. Each sub-dimension is scaled to [0, 1] using the minimum-maximum scaling method during offline processing.

[0077] Time difference characteristics ,in For the Event timestamps, is the starting point of the window. To prevent the scale from being too large and diluting the contribution of other dimensions, Perform zero mean unit variance standardization: , and are the mean and standard deviation of the time difference within the window respectively. Standardization ensures that the model maintains the comparability of time features under the conditions of early and late shifts, fast and slow loading rates. The node feature vector is obtained by splicing the three parts , all vectors are superimposed into a node feature matrix according to the event sequence This manual is unified with Represents a single node feature vector, Represents the node feature matrix. Then the graph ordinary differential equation is introduced. The convolution kernel adopts the GraphSAGE structure, and its aggregation function includes three steps: adjacency summation, linear mapping and normalization, which can be summarized as follows: ,in is the node hidden vector, For nodes The adjacency set of 、 are learnable weights, is a nonlinear activation. Adjacency sets are automatically derived from the nonzero indices of a third-order sparse tensor, eliminating the need for explicit graph construction. The GraphSAGE convolution kernel is embedded in the ordinary differential framework, forming the aforementioned time-continuous evolution equation. The Euclidean distance between the integrated hidden state matrix and the input feature matrix reflects the degree of agreement between the experimental behavior and the topological prior.

[0078] If the consistency score is lower than the threshold, the system performs two-level topological feedback: first, the element representing the one-dimensional ring structure in the topological barcode vector is multiplied by the gain coefficient , and then synchronously amplify the corresponding hyperedge weight in the third-order sparse tensor The logic behind the gain operation is to increase the influence of loop structure anomalies on subsequent window convolution aggregations when they are significant, enabling the model to detect similar violations more quickly. All gain details are written to the adjustment log, which includes fields such as the window timestamp, the index of the dimension affected by the gain, and the coefficient value.

[0079] The consistency score and adjustment log are binary serialized and written into a third data packet, with a Trusted Execution Environment elliptic curve signature appended to the end. Since the score only takes up 4 floating-point bytes, the log size averages less than 1kB. Combined with the 96-byte signature, the third data packet is lightweight and adaptable to limited bandwidth on-site.

[0080] Example verification: In a concrete shear and compression loading test, the window length is set to 200ms, with a total of 300 windows. The average score of the normal window is 0.97. After the 120th window, the insertion of a false device number event causes the score to drop to 0.63, which is lower than the threshold of 0.9. The system executes The third packet size was 14kB. The score for the subsequent three windows rose back to 0.92. Upon encountering a similar violation, the score immediately dropped to 0.45, triggering an alarm. Compared to the linear autoregressive baseline, the average response latency was reduced by 60% and the missed detection rate was reduced by 42%.

[0081] In summary, through the time difference normalization of the node feature matrix, topological injection of the GraphSAGE convolution kernel, and gain feedback, the present invention realizes continuous time monitoring of the behavior of mechanical test events. The third data packet solidifies the scoring and adjustment into verifiable credentials, providing a solid foundation for subsequent zero-knowledge auditing and responsibility tracing.

[0082] Preferably, when the consistency score is lower than the threshold, the element representing the one-dimensional ring structure in the topological barcode vector is multiplied by a gain coefficient, and the hyperedge weight associated with the element is adjusted by the same coefficient to generate an adjustment log.

[0083] Output consistency scores in continuous-time graphical models Afterwards, the present invention sets the threshold .like , indicating that the event behavior in the current window deviates significantly from the baseline topology, and the sensitivity of the next window to similar anomalies needs to be increased immediately. This invention proposes a "ring-hyperedge linkage gain" mechanism: select the topology barcode vector One-dimensional ring elements in As a regulatory target, the gain coefficient Synchronize the amplification loop importance and associated hyperedge weight. Its core calculation expression is: ; ,in is the post-gain loop count, represents the original weight of the hyperedge corresponding to the ring, is the adjusted weight. refers to the one-dimensional ring count in the topological barcode vector, The meaning is fixed within the same window. After the gain, the weight of the ring structure in the next window graph convolution aggregation is amplified, making the latent vector of the relevant node more sensitive to the same source abnormal combination, realizing the online "magnifying glass" effect.

[0084] Adjust the log to record the following fields: window timestamp, gain front , after gain , the number of participating hyperedges, coefficient and trigger ratings The log is binary serialized and takes up no more than 1kB. The log is written into the third data packet payload area; the index area stores the unique identifier The tail is signed by an elliptic curve generated by the trusted execution environment, and any node can use the public key to verify the integrity of the package once.

[0085] The one-dimensional loop structure corresponds to the closed dependency chain of the three elements of "man-machine-environment" in the test process. If an unlicensed person is replaced or the equipment is disconnected, part of the loop will be destroyed, causing A sharp drop. Through the gain Its hyperedges can apply external forces to the integral equation, causing the model to allocate more attention to the affected nodes and achieve rapid convergence. If the anomaly persists, the score is further reduced and the gain is repeated, forming an adaptive multi-stage amplification. If the anomaly disappears and the score recovers, the gain operation is naturally stopped to avoid false alarms.

[0086] Example: Bridge concrete shear and compression loading, window length 200ms, baseline threshold Normal stage The average value is 0.97. The 150th window inserts a fake device number event, and the score drops to 0.63. ,Will The score increased from 11 to 13.2, with the weights of 45 hyperedges simultaneously amplified. The third data packet was 14kB in size and written to the chain within 30ms. The score in the subsequent two windows rebounded to 0.91. If a counterfeit device or person skipped frames again, the score in the first window would drop to 0.48, triggering an on-chain alarm. Fifty consecutive rounds of random injection tests showed that the introduction of the gain reduced detection latency by an average of 60%, and the missed detection rate dropped from 0.08 to 0.04.

[0087] Using the third-order sparse tensor, topological barcode vector, consensus signature, consistency score and adjustment log as private input, generate and aggregate zero-knowledge proof and submit it to the distributed ledger; after verification by the audit node, write the electronic qualification certificate or record the violation information; integrate the first data packet, the second data packet, the third data packet and the aggregated proof to generate a traceability report and provide query.

[0088] In this stage, the third-order sparse tensor is aggregated through the zero-knowledge proof mechanism. , topological barcode vector , consensus signature , consistency score And adjust the log As private input, a mathematical proof is generated in a trusted execution environment and written to the distributed ledger. This four-step closed loop of proof-verification-credential-traceability allows for public proof of compliance to any node without leaking the information.

[0089] The proof circuit contains three constraints: 1. Signature integrity constraint, using the witness network public key set to With message digest Perform elliptic curve verification; 2. Topological consistency constraints, recalculated within the circuit , the check result is consistent with the private input; 3. Score correctness constraint, Substitute into the consistency function and verify .

[0090] The circuit uses the Groth16 protocol to output a single proof , the size is about 1200 bytes, and the verification takes about 12 milliseconds. To reduce the load on the chain, 100 window-level proofs are aggregated into 1 aggregate proof on the same GPU. Aggregation complexity , on-chain verification complexity , verification time is about 18 milliseconds. The private input hash is written as the public input proof. External nodes can verify it only with the public key and hash, without having access to the plaintext content.

[0091] The aggregation proof is submitted to the ledger along with the transaction. The ledger adopts the Byzantine fault-tolerant consensus of the authoritative node and produces a new block every 5 seconds. After a transaction is completed, if verification succeeds, an electronic certificate is generated. If it fails, the fingerprint model is invoked to recalculate the unique identifier and compare it with the on-chain hash. If there is a discrepancy, a violation is recorded. The certificate or violation entry fields include a timestamp, a unique identifier hash, and a block height, all of which are written to the chain to prevent rollbacks.

[0092] The traceability report service accepts the unique identifier hash query and sequentially pulls: the original text and hash of the first data packet, the second data packet, and the third data packet; the aggregation proof in the block and audit results; finally verify the package signature, and if they are consistent, combine the timeline, key hash, topology summary, scoring curve and audit conclusion to generate a structured report The report can be exported as PDF or JSON, and a block link is provided in the interface.

[0093] Through this invention, zero-knowledge proof hides 、 、 、 Audit nodes verify correctness solely through hashing. Only one 16kB aggregate proof is submitted for every 100 windows, reducing bandwidth by approximately 87% compared to a single 120kB proof uploaded directly to the chain. Verification failures can be pinpointed to the specific window index and hyperedge number, combined with fingerprint data to reconstruct the sample's identity, enabling minute-by-minute accountability. Proof generation, on-chain verification, credential writing, and report export are fully automated, reducing manual review cycles by 70%.

[0094] Example: Highway bridge batch contains 300 windows, cumulative 300 copies Aggregate into 3 parts Write chain, all verifications passed and qualified certificates were generated. Then manually tampered with the topological barcode vector of window index 145 to 147, and generated again The result of the circuit recalculation did not match, and the aggregate verification failed. The ledger immediately recorded the violation, and the traceability report automatically marked the abnormal window list, guiding the review and verification of the fingerprint sequence. Ultimately, it was determined that the illegal operation was performed by an unlicensed individual, and the test data was invalidated.

[0095] In summary, by encapsulating the full amount of structure-behavior-assessment data into zero-knowledge aggregate proof and writing it on-chain, the present invention ensures commercial confidentiality while achieving public verifiability, real-time auditing, and rapid accountability for the authenticity traceability of mechanical tests, significantly improving the transparency and reliability of transportation engineering quality reviews.

[0096] Preferably, the Groth16 proof scheme is adopted when generating the zero-knowledge proof, and the third-order sparse tensor, topological barcode vector, consensus signature, consistency score and adjustment log are used as private inputs, and the hash value of the above private inputs is used as public input.

[0097] The present invention uses Groth16 zero-knowledge proof in the chain release link to convert the third-order sparse tensor , topological barcode vector , consensus signature , consistency score And adjust the log As private input, concatenate the 256-bit hash values ​​of the five private data items into public input The proof circuit verifies the following three points through constraints: (1) Using the witness network public key set to With message hash Perform an elliptic curve pairing operation; (2) recalculate within the circuit 、 、 and with Compare; (3) Substitute into the consistency function and verify .in is the polynomial mapping obtained through offline training, is a tolerance constant. If all three constraints are met, the proof system outputs a single proof , the file size is about 1200 bytes.

[0098] In order to reduce the bandwidth on the chain, the present invention combines 100 window-level Compressed into 1 aggregate proof Aggregation is based on elliptic curve homomorphism, and the pairing equation is available: Perform a single verification. represents the bilinear pairing operation, 、 are the two endpoint elements of the aggregate proof, is the public input checkpoint, The verification cost is one pairwise multiplication, which takes approximately 18 milliseconds. The aggregated proof size is 16kB, saving approximately 87% of the bandwidth required for a single proof directly uploaded to the chain, which is 120kB.

[0099] The transaction contains the aggregation proof Hash with unique identifier , the new block is written within 5 seconds after the Byzantine fault-tolerant consensus of the authoritative node. The audit node automatically calls the validator for verification Successful verification will result in an electronic certificate containing fields such as a timestamp, unique identifier hash, and block height. Failure to verify the certificate is compared against the fingerprint hash, and if the difference exceeds a threshold, a violation is recorded. Neither the certificate nor the violation entry can be rolled back, ensuring post-hoc accountability.

[0100] The traceability report interface queryTrace(H) pulls the original text, aggregation certificate and audit results of the first to third data packets in sequence, verifies the signature at the end of the packet and generates a structured report The report concatenates the timeline, key hash, topology summary, consistency score curve, and audit conclusion into a hash chain, which can be exported as PDF or JSON.

[0101] Through this invention, only five hashes are disclosed on the chain and cannot be recovered externally. or After aggregation, the single verification maintains a constant level, which is suitable for field nodes with limited bandwidth. Any modification of private data will destroy the circuit constraints, resulting in Verification failed. A qualified certificate is generated upon successful verification, reducing report issuance time by approximately 70%. In the event of a verification failure, the report locates the specific window index and loop number, providing a basis for on-site audits.

[0102] Example: A batch of mechanical tests has 300 windows, generating 300 copies Aggregate into 3 parts The on-chain verification pass rate is 100%, and the electronic qualification certificate is written with the block. Then the topological barcode vector of window 145 to 147 is deliberately tampered with, making The sparse tensor was added by 1 but not modified synchronously. After regenerating the single proof, constraint 2 was not satisfied, the aggregate verification failed, and the ledger record was in violation. The traceability report automatically marked the abnormal window in red, guiding the reviewer to use the fingerprint scan to verify the sample. Ultimately, the operation was confirmed to be performed by an unlicensed individual and the test results were canceled. This example demonstrates that the zero-knowledge aggregate proof scheme of the present invention can reveal data tampering in real time without leaking confidentiality, significantly improving the transparency of the authenticity traceability system for traffic engineering mechanics tests.

[0103] Preferably, Figure 2 As shown in the figure, multiple zero-knowledge proofs are aggregated into a single aggregate proof through the homomorphic properties of elliptic curves, and a signature is generated by a trusted execution environment before being submitted to the distributed ledger.

[0104] The goal of this phase is to aggregate continuously generated zero-knowledge single proofs into a publicly verifiable aggregate proof of constant size without leaking any window-level private data, and to ensure that the aggregate proof is integrity signed by a trusted execution environment before being written to the distributed ledger.

[0105] Groth16 single proof consists of triples All three are group elements on the elliptic curve. Let the curve generator be , the bilinear pairing operator is , the verification equation can be simplified to: ,in is the public verification key. Since group operations are homomorphic, the corresponding coordinates of different proofs can be added element by element on the curve. Let the window index The single proof is ,right The proof is summed coordinate by coordinate to get: , after aggregation, the verification condition is transformed into: , verification maintains the same computational complexity as single proof, and constant-level verification can be achieved with 111 pairings. This property is known as "elliptic curve homomorphic aggregation."

[0106] The implementation process includes: window-level proof generation, trusted execution environment private input for each window Run the Groth16 prover and output a single proof .

[0107] Homomorphic aggregation, when the cumulative number of windows reaches the threshold When the coordinates are added in the same execution environment, The aggregation process does not require access to the original private key or public verification key to prevent leakage midway.

[0108] Aggregate proof signature, after aggregation is completed, the execution environment calls the hardware key pair Generate elliptic curve digital signature .in The hash is the unique identifier of the sample. Written together with the ledger transaction payload.

[0109] Ledger verification, the distributed ledger node first verifies with the review public key , and then use the public verification key to Perform a pairing check. The block can be placed only after both steps pass.

[0110] In this example, the bridge shear-compression loading batch has 300 windows, and 1 certificate is aggregated for every 100 windows, generating a total of 3 certificates. . All on-chain verifications passed and three electronic qualification certificates were automatically written. Subsequently, the topological barcode vectors of windows 145-147 were deliberately tampered with and re-aggregated. Due to inconsistent constraints, the verification failed, the ledger generated a violation entry and rejected the transaction. The traceability report API locates the abnormal window based on the unique identifier hash, prompting the reviewer to review the original fingerprint data for verification, and finally confirmed that the laboratory was operated by unlicensed personnel during that time period, and the test results were invalidated. The experimental results show that homomorphic aggregation combined with trusted execution environment signatures can quickly expose data tampering while ensuring controllable on-chain load, providing an efficient and robust blockchain evidence storage solution for authenticity traceability of traffic engineering mechanics tests.

[0111] Preferably, the timestamp of the event stream is generated by a time synchronization system compliant with the IEEE1588 protocol and written along with the event.

[0112] This method requires that every event in the mechanical testing process be timestamped with high precision and global consistency, ensuring strict causal order when constructing a third-order sparse tensor from multiple data sources. To achieve this, the system deploys a time synchronization network compliant with IEEE 1588-2008 (Precision Time Protocol Version 2, abbreviated in the industry as PTPv2).

[0113] IEEE1588 adopts a "master-slave model". The atomic clock of the synchronous satellite signal outside the laboratory acts as the master clock, and periodically broadcasts four types of messages, Sync, Follow_Up, Delay_Req, and Delay_Resp, to all test nodes via Gigabit Ethernet. The node network card hardware quickly captures the arrival time of the message and calculates the instantaneous deviation between the local clock and the master clock. If the real time of the slave node is , the main clock time is , then the synchronization error is defined as ,in Refers to the absolute time difference after a single calibration. Experimental measurements show that in a Gigabit copper cable topology and single-hop scenario, ≤200ns, meeting the 1µs time resolution requirement for parallel testing of multiple loaders.

[0114] The network topology uses a master clock that extends to the test area via two levels of boundary clocks. Terminals such as the loader, environmental chamber, and fluorescence camera are equipped with network interface controllers that support hardware timestamping. The network utilizes a dedicated VLAN, disabling Energy Efficient Ethernet (EEE) to minimize jitter in message queues.

[0115] In the timestamp injection process, the acquisition driver reads the network card hardware counter in the DMA link and writes the absolute UTC second and nanosecond fields into the event header. After the upper-layer acquisition thread obtains the five-element status, it immediately concatenates the timestamp with the status bytecode and writes it into the ring buffer. The ring buffer is refreshed to the event stream file every 10Hz to ensure "data-time" atomic writes.

[0116] Drift monitoring and self-diagnosis, the master clock pushes an Announce message every 60 seconds, node detection Checks whether the 1µs alarm threshold is exceeded. If so, the node submits a syncWarning event, and subsequent sparse tensor construction will mark the record as a dishonest node.

[0117] This invention ensures that all events share the same UTC reference plane, ensuring a fully ordered relationship regardless of the order in which the loader or camera is activated, thus preventing fraudulent practices such as "backfilling data." During the third-order tensor construction phase, a three-event combination must have a "time difference ≤ 2ms." Asynchronous timestamps will directly invalidate the combination, exposing cross-node forgeries. The consistency scoring process uses time difference as a node feature, improving accuracy and enabling the model to discern loading rate fluctuations as low as 0.1Hz.

[0118] Example: In a shear-compression loading test on a bridge, the window length is 200ms, and 36,000 events are written in 300 consecutive windows. The average value is 110ns, the maximum value is 430ns, and no single value exceeds the 1µs warning line. The timestamp function of the network port of a loader was manually disabled, resulting in the node The time instantaneously increased to 15µs, which was captured by the syncWarning event. During the consistency scoring phase, the score dropped to 0.58 (threshold 0.90), subsequently triggering topology gain and audit alerts. After review, it was confirmed that the node's network card had been replaced with a model that did not support PTP, and the test data was marked invalid.

[0119] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware.

[0120] The above are merely embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should be included within the scope of the claims of the present application.

Claims

1. A mechanical test authenticity tracing method, characterized in that: include: Apply a unique identifier to the sample and simultaneously obtain 3D measurement data to generate fingerprint data and 3D model; hash the two to obtain a unique identifier and perform erasure coding; Collecting personnel, equipment, process, environment and sample status information under a unified time base to form an event stream and encapsulate it into a first data packet; Map the event stream into a third-order sparse tensor within the set window, extract the topological barcode vector, generate a consensus signature using the threshold signature, and encapsulate it with the unique identifier into a second data packet; Solving a graph-based ordinary differential equation containing topological constraints based on the third-order sparse tensor and event stream to obtain a consistency score; if the consistency score is lower than a threshold, adjusting the topological barcode vector and hyperedge weight to generate an adjustment log, and encapsulating the log and the consistency score into a third data packet; Using the third-order sparse tensor, topological barcode vector, consensus signature, consistency score and adjustment log as private input, generate and aggregate zero-knowledge proof and submit it to the distributed ledger; after verification by the audit node, write the electronic qualification certificate or record the violation information; integrate the first data packet, the second data packet, the third data packet and the aggregated proof to generate a traceability report and provide query.

2. The method according to claim 1, characterized in that When applying a unique identification to the sample, nucleic acid barcode particles are deposited in a spray manner, and the nucleic acid barcode particles produce a unique fluorescence response in the ultraviolet band; the sample fingerprint data and the sample three-dimensional model data are recorded with the same timestamp and spatial coordinates before packaging.

3. The method according to claim 1, characterized in that When mapping the event stream into a third-order sparse tensor, a fixed-length time window is used to sort the events, and three event records with a time difference no greater than a preset threshold and a consistent source are written into the third-order sparse tensor as a tensor index.

4. The method according to claim 3, characterized in that When extracting the topological barcode vector, a Vietoris–Rips complex is constructed for the third-order sparse tensor, and zero-dimensional, one-dimensional, and two-dimensional topological invariants are calculated and combined in the order of appearance to form a topological barcode vector.

5. The method according to claim 4, characterized in that The threshold signature algorithm is applied when generating the consensus signature. The threshold value is equal to the total number of witness nodes minus the number of tolerable failure nodes plus one.

6. The method according to claim 1, wherein Before solving the graph neural network ordinary differential equation, a node feature vector is constructed for each event based on the third-order sparse tensor and event stream data, and the time difference features are standardized to form a node feature matrix; the graph convolution kernel adopts the GraphSAGE structure.

7. The method according to claim 6, characterized in that When the consistency score is lower than the threshold, the element representing the one-dimensional ring structure in the topological barcode vector is multiplied by a gain coefficient, and the hyperedge weight associated with the element is adjusted by the same coefficient to generate an adjustment log.

8. The method according to claim 1, characterized in that The Groth16 proof scheme is adopted when generating the zero-knowledge proof, and the third-order sparse tensor, topological barcode vector, consensus signature, consistency score and adjustment log are used as private inputs, and the hash value of the above private inputs is used as public input.

9. The method according to claim 8, characterized in that Multiple zero-knowledge proofs are aggregated into a single aggregate proof through elliptic curve homomorphism, and a signature is generated by a trusted execution environment before being submitted to the distributed ledger.

10. The method according to claim 1, characterized in that The timestamps of event streams are generated by a time synchronization system that complies with the IEEE1588 protocol and written along with the events.

Images