Device predictive maintenance method and system based on block chain

Abstract

The invention provides an equipment predictive maintenance method and system based on a block chain, and belongs to the field of data analysis and data mining, and the method comprises the steps: S1, obtaining original data through a sensor or other trusted modes; s2, a distributed architecture is designed, all participants finish training locally, and original data does not need to be exposed; s3, realizing parameter interaction between each participant and the central server through the constructed encryption mechanism; s4, constructing a bottom-layer architecture based on a block chain, verifying the data and recording parameter changes in a training process; and S5, designing an incentive mechanism, specifying a contribution calculation mode of each participant to the model, and issuing awards. According to the method, cross-organization predictive maintenance framework construction can be realized, data value sharing is realized under the condition of protecting data privacy, the production efficiency is improved, and intelligence in modern industrial production is promoted.

Description

Technical Field

[0001] The present invention belongs to the technical field of data analysis and data mining, and in particular relates to a blockchain-based equipment predictive maintenance method and system. Background Art

[0002] Blockchain is a distributed ledger technology that records transactions or data in a decentralized manner to ensure transparency, immutability and security of information. Each block contains multiple records, which are linked into a chain and verified through a consensus mechanism to prevent data tampering. It effectively solves the problem of privacy protection of predictive maintenance data and can achieve multi-party collaborative computing without circulating original data to ensure data security and reliability. This makes blockchain an ideal choice for multiple participants to jointly train predictive maintenance models.

[0003] Federated learning is a distributed machine learning method that allows multiple parties to jointly train a global model without sharing local data. After training locally, each party only provides the training results for model updates. In predictive maintenance, federated learning allows different organizations to collaboratively train equipment predictive maintenance models without exposing local equipment data, which improves the accuracy and versatility of the model.

[0004] As an important link in the industrial chain supply chain, the operation and maintenance of manufacturing equipment is an important factor affecting production capacity and efficiency. It is of vital importance to ensure the smooth production process and improve production efficiency. Accurate predictive maintenance helps reduce production interruptions caused by equipment damage. At present, predictive maintenance mainly uses expert knowledge and traditional machine learning methods based on a small amount of local data, and the prediction effect is limited. At the same time, in order to protect local data security, enterprises are reluctant to share data resources, which leads to the inability to fully realize the value of predictive maintenance data. Therefore, how to use blockchain and federated learning technology to perform predictive maintenance on equipment has become an urgent problem to be solved. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a blockchain-based equipment predictive maintenance method, comprising the following steps: Step S1: Obtaining raw data of device operation through sensors or other trusted methods; Step S2: constructing a global predictive maintenance model, using the original data to perform local model training on multiple participants using a federated learning algorithm, uploading the training results of each participant to a central server for aggregation and updating the global predictive maintenance model, and sending the updated global predictive maintenance model parameters to each participant for local model training again, until a trained global predictive maintenance model and local models are obtained; Step S3: construct an encryption mechanism based on a homomorphic encryption algorithm, introduce a multi-party collaborative computing mechanism into the global predictive maintenance model and the local model to achieve data privacy protection; Step S4: constructing an underlying architecture based on blockchain to complete collaborative computing among all participants in a decentralized manner without circulating the original data; Step S5: Design an incentive mechanism to calculate the contribution value of each participant and obtain the model proportion of each participant.

[0006] Beneficial effects: 1. The method of the present invention uses blockchain as the underlying architecture, completes multi-party collaborative computing in a decentralized manner without circulating original data, and realizes data sharing on the basis of ensuring data security. This architecture ensures that the process of joint training of predictive maintenance models by all participants is credible and traceable.

[0007] 2. The present invention constructs an encryption mechanism based on the homomorphic encryption algorithm to encrypt the training results of different terminals, ensuring that the central server can integrate and update the global model while avoiding the leakage of model parameters.

[0008] 3. The present invention designs a distributed architecture and adopts a federated learning algorithm to implement distributed training. Each participant uses local resources for training without exposing the original data, thus protecting data privacy and security.

[0009] 4. The present invention designs an automatic reward distribution mechanism based on smart contracts that takes into account the contribution of the global model. In the scenario of distributed training, it stipulates the contribution calculation method of different participants, thereby defining the model proportion of each participant, and designs a new reward method based on smart contracts to increase the enthusiasm of each participant to join model training and improve the quality of model training data. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 This is a flowchart of a blockchain-based equipment predictive maintenance method of the present invention; Figure 2 Schematic diagram of the data flow process for blockchain integrating federated learning and homomorphic encryption; Figure 3 It is a schematic diagram of the framework structure of the method of the present invention; Figure 4 This is a structural block diagram of a blockchain-based equipment predictive maintenance system of the present invention. DETAILED DESCRIPTION

[0011] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0012] Embodiment 1 like Figure 1 As shown, an embodiment of the present invention provides a blockchain-based equipment predictive maintenance method, comprising the following steps: Step S1: Obtaining raw data of device operation through sensors or other trusted methods; Step S2: construct a global predictive maintenance model, use the federated learning algorithm to train local models for multiple participants using the original data, upload the training results of each participant to the central server for aggregation and update the global predictive maintenance model, and send the updated global predictive maintenance model parameters to each participant for local model training again, until a trained global predictive maintenance model and local models are obtained; Step S3: construct an encryption mechanism based on the homomorphic encryption algorithm, introduce a multi-party collaborative computing mechanism into the global predictive maintenance model and the local model to achieve data privacy protection; Step S4: Build an underlying architecture based on blockchain to complete collaborative computing among all participants in a decentralized manner without circulating the original data; Step S5: Design an incentive mechanism to calculate the contribution value of each participant and obtain the model proportion of each participant.

[0013] In one embodiment, the above step S1: obtaining the original data of the device operation through a sensor or other trusted method specifically includes: 1) Static data, collect relevant parameter data and maintenance records of the equipment, including but not limited to the rated power, rated speed, model structure and equipment failure records of the equipment; 2) Dynamic data: collect parameter data and environmental data during equipment operation through sensors, including but not limited to current, voltage, speed, temperature and humidity; 3) System data: data related to equipment events in production-related systems, including but not limited to abnormal events of equipment during the production scheduling process of the MES system.

[0014] In one embodiment, the above step S2: constructing a global predictive maintenance model, using a federated learning algorithm to perform local model training on multiple participants using original data, uploading the training results of each participant to a central server for aggregation and updating the global predictive maintenance model, and sending the updated global predictive maintenance model parameters to each participant for local model training again, until a trained global predictive maintenance model and local models are obtained, specifically includes: Step S21: Connect the devices or systems of multiple participants to the federated learning platform, where each participant retains control over the data maintained on its local device and establishes a secure connection with the central server through an encrypted communication protocol; Step S22: Each participant pre-processes the local equipment maintenance data, and the pre-processed data will be used for model training; Step S23: Initializing a global predictive maintenance model in the central server; Step S24: the central server sends the initialized global predictive maintenance model to each participant; Step S25: Each participant trains the received global predictive maintenance model locally to generate model weights or gradient information for updating the local predictive maintenance model; Step S26: Each participant uploads the model results of local training to the central server; Step S27: The central server receives and aggregates the local models from all participants and updates the global predictive maintenance model; Step S28: The central server distributes the updated global predictive maintenance model to each participant again, and repeats steps S25 to S28 until the preset result is achieved.

[0015] In one embodiment, the above step S3: constructing an encryption mechanism based on a homomorphic encryption algorithm, introducing a multi-party collaborative computing mechanism in the global predictive maintenance model and the local model to achieve data privacy protection, specifically includes: Step S31: After each participant completes the local model training, the result is encrypted using the homomorphic encryption algorithm. Assume that the training model parameter of participant i is the vector , using the public key right After encryption, the model parameters will be encrypted Upload to the central server, where the encryption formula is as follows: (1) in, is the encryption function of the homomorphic encryption algorithm; Step S32: The central server receives the encrypted model parameters uploaded by each participant. After that, the homomorphic aggregation operation is directly performed without decryption. The aggregation operation is shown in formula (2), which converts the complete global predictive maintenance model parameters into Sent to each party in an encrypted state; (2) in, is the weight of each participant model, is the number of parties involved in the training; Step S33: Each participant receives and uses the corresponding private key right Decryption is performed as shown in formula (3); after decryption, each participant can use the global predictive maintenance model parameters for reasoning or continue local training.

[0016] )) (3) in, is the decryption function of the homomorphic encryption algorithm.

[0017] In one embodiment, the above step S4: constructing an underlying architecture based on blockchain to complete the collaborative computing of all participants in a decentralized manner without circulating the original data, specifically includes: Step S41: The central server verifies the local data of each participant and uses an improved hierarchical timestamp mechanism to bind the original source of the device sensor data, the device identification and the corresponding smart contract hash value to the same block; where the timestamp The calculation formula is as follows: (4) in, Yes Equipment In time The data generated, is the unique identifier of the device, To operate the contract; The timestamps of all devices are integrated through a hierarchical hash tree, and the calculation formula is as follows: ( || ||...|| ) (5) In this step, the local data of each participant is verified through the central server to ensure data transparency and non-tamperability. The layered timestamp mechanism can efficiently detect data consistency.

[0018] Step S42: A sharing mechanism based on multi-party secure computing and zero-knowledge proof is adopted, allowing all participants to share and verify predictive maintenance data without exposing sensitive data, completing cross-organizational data collaboration, thereby achieving multi-party data sharing and collaboration; Step S43: Using smart contracts to define and automatically execute predictive maintenance protocols, and using an adaptive anomaly detection module to perform real-time detection, automatically arranging equipment inspections and notifying maintenance personnel when an anomaly is detected; Adaptive anomaly detection module uses historical data from local devices and real-time data , through a blockchain-based machine learning model Make dynamic adjustments to perform real-time detection, reduce manual intervention and improve response speed.

[0019] Figure 2 A schematic diagram of the data flow process of blockchain integrating federated learning and homomorphic encryption is shown.

[0020] In one embodiment, the above step S5: designing an incentive mechanism to calculate the contribution value of each participant and obtain the model proportion of each participant specifically includes: Step S51: Calculate the participants The contribution ratio of the size of the local training dataset , Data quality contribution ratio And the contribution ratio to the global predictive maintenance model optimization , the calculation formula is as follows; (6) (7) (8) (9) in, For the participants The size of the local training dataset; is the total dataset size of all participants; The data quality depends on factors such as noise, duplication, and label accuracy of each participant; For the participants Optimization contribution to the global predictive maintenance model; is the t-1th round loss function of the global predictive maintenance model, For the participants The loss function of the global predictive maintenance model after the tth round of local model update; Set weight ratio , , Calculate the weight ratio of the participants corresponding to the size of the data set, data quality and optimization contribution Total weight The formula is as follows: (10) Smart contracts based on For participants Issuing rewards or recording model proportions: (11) in, is the total reward set by the smart contract; Step S52: Based on the changes in model parameters during the training process of each participant recorded in the blockchain, calculate the contribution of the global predictive maintenance model: Assume that in In rounds, participants The global predictive maintenance model parameters before the local model is updated are , the difference after the local model parameters are updated Defined as: (12) in, Indicates the participants In the The contribution of local model training in a round to the update of global predictive maintenance model parameters; After each round of iteration, the global predictive maintenance model parameters are updated by taking a weighted average of the local model parameter changes of each participant. : (13) Due to the difference in the model parameter updates of each participant in each round of iteration , the contribution value of each participant All will be recorded in the blockchain. The present invention designs gradient clipping, gradient consistency detection and loss function check to check the rationality of the model: Among them, the gradient clipping formula is as follows: (14) in, is the gradient clipping function, A threshold for the update amplitude of the global predictive maintenance model parameters; The gradient consistency formula is as follows: (15) in, is the gradient consistency function, It is the similarity threshold of the updates of the global predictive maintenance model by two different participants in the current round of updates. If the update of one participant deviates significantly from that of other participants, it will be automatically marked as suspicious and the update will be rejected. By limiting the amplitude of model updates, malicious participants are prevented from exaggerating local training effects and affecting the global model. The gradient consistency proposed in this invention can detect abnormal updates and mark them by comparing the similarities between model updates uploaded by each participant. The loss function check determines whether the model update result of the participant will lead to an increase in the global model loss function to determine whether it is a normal update. The loss function check formula is as follows: (16) Formula (16) means that when the participants When the model update does not bring actual optimization effect, the update is automatically deemed invalid.

[0021] Figure 3 A schematic diagram of the framework structure of the method of the present invention is shown.

[0022] Embodiment 2 like Figure 4 As shown, an embodiment of the present invention provides a blockchain-based equipment predictive maintenance system, including the following modules: The data acquisition module 61 is used to acquire the original data of the device operation through sensors or other trusted methods; The model training module 62 is used to build a global predictive maintenance model, use the original data of the federated learning algorithm to perform local model training on multiple participants, upload the training results of each participant to the central server for aggregation and update the global predictive maintenance model, and send the updated global predictive maintenance model parameters to each participant for local model training again, until a trained global predictive maintenance model and local models are obtained; An encryption module 63 is used to build an encryption mechanism based on a homomorphic encryption algorithm, introduce a multi-party collaborative computing mechanism into a global predictive maintenance model and a local model, and realize data privacy protection; The collaborative computing module 64 is used to build an underlying architecture based on blockchain, and complete collaborative computing among all participants in a decentralized manner without circulating original data; The incentive mechanism module 65 is used to design an incentive mechanism to calculate the contribution value of each participant and obtain the model proportion of each participant.

Claims

1. A blockchain-based equipment predictive maintenance method, characterized in that: include: Step S1: Obtaining raw data of device operation through sensors or other trusted methods; Step S2: constructing a global predictive maintenance model, using the original data to perform local model training on multiple participants using a federated learning algorithm, uploading the training results of each participant to a central server for aggregation and updating the global predictive maintenance model, and sending the updated global predictive maintenance model parameters to each participant for local model training again, until a trained global predictive maintenance model and local models are obtained; Step S3: construct an encryption mechanism based on a homomorphic encryption algorithm, introduce a multi-party collaborative computing mechanism into the global predictive maintenance model and the local model to achieve data privacy protection; Step S4: constructing an underlying architecture based on blockchain to complete collaborative computing among all participants in a decentralized manner without circulating the original data; Step S5: Design an incentive mechanism to calculate the contribution value of each participant and obtain the model proportion of each participant.

2. The device predictive maintenance method based on blockchain according to claim 1 is characterized in that: The step S1: obtaining the original data of the device operation through sensors or other reliable methods, specifically includes: 1) Static data, collecting relevant parameter data and maintenance records of equipment; 2) Dynamic data: collect parameter data and environmental data during equipment operation through sensors; 3) System data: data related to equipment events in production-related systems.

3. The device predictive maintenance method based on blockchain according to claim 2 is characterized in that: The step S2: constructing a global predictive maintenance model, using the original data to perform local model training on multiple participants using a federated learning algorithm, uploading the training results of each participant to a central server for aggregation and updating the global predictive maintenance model, and sending the updated global predictive maintenance model parameters to each participant for local model training again, until a trained global predictive maintenance model and local models are obtained, specifically including: Step S21: Connecting the devices or systems of multiple participants to the federated learning platform, each participant retains control over the data maintained by its local device and securely connects to the central server through an encrypted communication protocol; Step S22: Each participant pre-processes the local equipment maintenance data, and the pre-processed data will be used for model training; Step S23: Initializing a global predictive maintenance model in the central server; Step S24: the central server sends the initialized global predictive maintenance model to each participant; Step S25: Each participant trains the received global predictive maintenance model locally to generate model weights or gradient information for updating the local predictive maintenance model; Step S26: Each participant uploads the model results of local training to the central server; Step S27: the central server receives and aggregates the local models from all participants and updates the global predictive maintenance model; Step S28: The central server distributes the updated global predictive maintenance model to each participant again, and repeats steps S25 to S28 until the preset result is achieved.

4. The device predictive maintenance method based on blockchain according to claim 3 is characterized in that: The step S3: constructing an encryption mechanism based on a homomorphic encryption algorithm, introducing a multi-party collaborative computing mechanism into the global predictive maintenance model and the local model to achieve data privacy protection, specifically including: Step S31: After each participant completes the local model training, the result is encrypted using the homomorphic encryption algorithm. Assume that the training model parameter of participant i is the vector , using a public key right After encryption, the model parameters will be encrypted Upload to the central server, where the encryption formula is as follows: (1) in, is the encryption function of the homomorphic encryption algorithm; Step S32: The central server receives the encrypted model parameters uploaded by each participant. After that, the homomorphic aggregation operation is directly performed without decryption. The aggregation operation is shown in formula (2), which converts the complete global predictive maintenance model parameters into Sent to each party in an encrypted state; (2) in, is the weight of each participant model, is the number of parties involved in the training; Step S33: Each participant receives and uses the corresponding private key right Decryption is performed as shown in formula (3); after decryption, each participant can use the global predictive maintenance model parameters for reasoning or continue local training; )) (3) in, is the decryption function of the homomorphic encryption algorithm.

5. The device predictive maintenance method based on blockchain according to claim 4 is characterized in that: The step S4: constructing an underlying architecture based on blockchain, and completing the collaborative calculation of each participant in a decentralized manner on the basis of non-circulation of the original data, specifically including: Step S41: The central server verifies the local data of each participant and uses an improved hierarchical timestamp mechanism to bind the original source of the device sensor data, the device identification and the corresponding smart contract hash value to the same block; where the timestamp The calculation formula is as follows: (4) in, Yes Equipment In time The data generated, is the unique identifier of the device, To operate the contract; The timestamps of all devices are integrated through a hierarchical hash tree, and the calculation formula is as follows: ( || ||...|| ) (5) Step S42: A sharing mechanism based on multi-party secure computing and zero-knowledge proof is adopted, allowing all participants to share and verify predictive maintenance data without exposing sensitive data, thereby completing cross-organizational data collaboration; Step S43: Use smart contracts to define and automatically execute predictive maintenance protocols, perform real-time detection through an adaptive anomaly detection module, and automatically arrange equipment inspections and notify maintenance personnel when an anomaly is detected.

6. The blockchain-based equipment predictive maintenance method according to claim 5 is characterized in that: Step S5: designing an incentive mechanism to calculate the contribution value of each participant and obtain the model proportion of each participant, specifically including: Step S51: Calculate the participants The contribution ratio of the size of the local training dataset , Data quality contribution ratio And the contribution ratio to the global predictive maintenance model optimization , the calculation formula is as follows; (6) (7) (8) (9) in, For the participants The size of the local training dataset; is the total dataset size of all participants; for data quality; For the participants Optimization contribution to the global predictive maintenance model; is the t-1th round loss function of the global predictive maintenance model, For the participants The loss function of the global predictive maintenance model after the tth round of local model update; Set weight ratio , , Calculate the weight ratio of the participants corresponding to the size of the data set, data quality and optimization contribution Total weight The formula is as follows: (10) Smart contracts based on For the participants Issuing rewards or recording model proportions: (11) in, is the total reward set by the smart contract; Step S52: Based on the changes in model parameters during the training process of each participant recorded in the blockchain, calculate the contribution of the global predictive maintenance model: Assume that in In rounds, participants The global predictive maintenance model parameters before the local model is updated are , the difference after the local model parameters are updated Defined as: (12) in, Indicates the participants In the The contribution of local model training in a round to the update of global predictive maintenance model parameters; After each round of iteration, the global predictive maintenance model parameters are updated by taking a weighted average of the local model parameter changes of each participant. : (13) Use gradient clipping, gradient consistency detection and loss function check to check the rationality of the model: Among them, the gradient clipping formula is as follows: (14) in, is the gradient clipping function, A threshold for the update amplitude of the global predictive maintenance model parameters; The gradient consistency formula is as follows: (15) in, is the gradient consistency function, It is the similarity threshold of the updates of the global predictive maintenance model by two different participants in the current round of updates. If the update of one participant deviates significantly from that of other participants, it will be automatically marked as suspicious and the update will be rejected. The loss function check formula is as follows: (16) Formula (16) means that when the participants When the model update does not bring actual optimization effect, the update is automatically deemed invalid.

7. A blockchain-based equipment predictive maintenance system, characterized in that: Includes the following modules: The data acquisition module is used to obtain the original data of the device operation through sensors or other trusted methods; A model training module, which is used to build a global predictive maintenance model, use the raw data to perform local model training on multiple participants using a federated learning algorithm, upload the training results of each participant to a central server for aggregation and update of the global predictive maintenance model, and send the updated global predictive maintenance model parameters to each participant for local model training again, until a trained global predictive maintenance model and local models are obtained; An encryption module is used to build an encryption mechanism based on a homomorphic encryption algorithm, introduce a multi-party collaborative computing mechanism into the global predictive maintenance model and the local model, and realize data privacy protection; The collaborative computing module is used to build an underlying architecture based on blockchain, and complete the collaborative computing of all participants in a decentralized manner on the basis of non-circulation of the original data; The incentive mechanism module is used to design an incentive mechanism to calculate the contribution value of each participant and obtain the model proportion of each participant.

Images