Remaining useful life prediction method, device, equipment, medium and product

Abstract

The present disclosure relates to the technical field of reliability engineering, and particularly provides a remaining useful life prediction method, device, equipment, medium and product, comprising: obtaining sensor data of an industrial equipment; wherein the sensor data is used to indicate an operating state of the industrial equipment; pulse coding the sensor data to obtain a pulse sequence; wherein the pulse sequence is used to describe dynamic changes of the operating state at different time points; inputting the pulse sequence into a pulse neuron to determine a state parameter of the pulse neuron; and predicting a remaining useful life of the industrial equipment based on the state parameter. Through the above processing manner, the time sequence dependency relationship of the operating state can be effectively learned, and the fine-grained time sequence features in the pulse sequence can be mined, thereby improving the prediction accuracy and prediction accuracy of the remaining useful life.

Description

Technical Field

[0001] This disclosure relates to the technical field of reliability engineering, and more specifically, to a method, apparatus, device, medium, and product for predicting remaining useful life. Background Technology

[0002] Industrial equipment will experience performance degradation due to wear, fatigue, or environmental factors during long-term operation. If the remaining useful life (RUL) of industrial equipment cannot be predicted in time, it will cause the industrial equipment to shut down or even cause safety accidents.

[0003] In related technologies, traditional machine learning models and manual feature extraction are commonly used for RUL prediction. However, this approach cannot effectively learn the temporal dependency features of industrial equipment, resulting in insufficient prediction accuracy of RUL. Summary of the Invention

[0004] This disclosure is made in view of the above-mentioned problems. This disclosure provides a method, apparatus, device, medium, and product for predicting remaining useful life.

[0005] According to one aspect of this disclosure, a method for predicting remaining useful life is provided, comprising:

[0006] Acquire sensor data from industrial equipment; wherein the sensor data is used to indicate the operating status of the industrial equipment;

[0007] The sensor data is pulse encoded to obtain a pulse sequence; the pulse sequence is used to describe the dynamic changes of the operating state at different points in time.

[0008] The pulse sequence is input into the spiking neuron to determine the state parameters of the spiking neuron;

[0009] Based on the state parameters, the remaining service life of the industrial equipment is predicted.

[0010] Furthermore, according to one aspect of this disclosure, inputting the pulse sequence into a spiking neuron to determine the state parameters of the spiking neuron further includes:

[0011] The pulse sequence is mapped to a target vector space; wherein the target vector space is used to describe the membrane potential of the spiking neuron at different time points;

[0012] The total number of pulse releases by the spiking neuron is determined based on the potential changes of the membrane potential.

[0013] The membrane potential and the total number of pulse releases are used as the state parameters.

[0014] Furthermore, mapping the pulse sequence to the target vector space according to one aspect of this disclosure further includes:

[0015] Obtain the pulse release state of multiple spiking neurons at the first time point;

[0016] The pulse release state and the target pulse sequence are weighted and summed to obtain the target weighted result; wherein, the target pulse sequence is the pulse sequence at the first time point;

[0017] Determine the membrane potential decay rate of the spiking neuron at the first time point;

[0018] The target weighted result and the membrane potential decay rate are summed to obtain the membrane potential of the spiking neuron at the second time point; wherein the second time point is after the first time point.

[0019] Furthermore, according to one aspect of this disclosure, performing a weighted summation process on the pulse release state and the target pulse sequence to obtain a target weighted result further includes:

[0020] The first weight and the corresponding pulse release state are weighted to obtain a first weighted result; wherein, the first weight is used to indicate the connection strength between the spiking neuron and other spiking neurons;

[0021] The second weight and the corresponding target pulse sequence are weighted to obtain a second weighted result; wherein, the second weight is used to indicate the degree of influence of the target pulse sequence on the spiking neuron;

[0022] The first weighted result and the second weighted result are summed to obtain the target weighted result.

[0023] Furthermore, determining the membrane potential decay rate of the spiking neuron at the first time point according to one aspect of this disclosure further includes:

[0024] Determine the time difference between the first time point and the second time point;

[0025] The time difference and membrane time constant are subjected to an exponential operation to obtain the calculation result;

[0026] The product of the calculation result and the membrane potential of the spiking neuron at the first time point is determined as the membrane potential decay rate.

[0027] Furthermore, according to one aspect of this disclosure, pulse encoding of the sensor data to obtain a pulse sequence further includes:

[0028] Extract operational status indicators from the sensor data according to at least one extraction dimension;

[0029] Map the preprocessing results of the operating status indicators to the pulse transmission time;

[0030] The pulse sequence is generated based on the pulse transmission time.

[0031] Furthermore, mapping the preprocessing results of the operating status indicators to the pulse transmission time, according to one aspect of this disclosure, further includes:

[0032] The pulse transmission time is determined based on the preprocessing result and the mapping time; wherein, the mapping time is the time from the preprocessing result to the pulse transmission time.

[0033] Furthermore, predicting the remaining service life of the industrial equipment based on the state parameters according to one aspect of this disclosure further includes:

[0034] The membrane potential and total number of pulse releases of the spiking neurons are weighted and summed to obtain a third weighted result;

[0035] The remaining useful life is predicted based on the third weighted result.

[0036] Furthermore, according to one aspect of this disclosure, the remaining useful life prediction method further includes:

[0037] Differentiate the pulse release state function to obtain the derivative result;

[0038] Based on the product of the derivative result and the learning rate, the first weight and the second weight are optimized to obtain the optimized first weight and the optimized second weight; wherein, the learning rate is used to indicate the adjustment range of the first weight and the second weight;

[0039] The optimized first weight is used as the weight of the pulse release state, and the optimized second weight is used as the weight of the target pulse sequence.

[0040] According to another aspect of this disclosure, a remaining useful life prediction device is provided, comprising:

[0041] An acquisition unit is used to acquire sensor data from industrial equipment; wherein the sensor data is used to indicate the operating status of the industrial equipment.

[0042] A pulse coding unit is used to perform pulse coding on the sensor data to obtain a pulse sequence; wherein, the pulse sequence is used to describe the dynamic changes of the operating state at different time points;

[0043] A parameter determination unit is used to input the pulse sequence into the spiking neuron in order to determine the state parameters of the spiking neuron;

[0044] A prediction unit is used to predict the remaining service life of the industrial equipment based on the state parameters.

[0045] According to another aspect of this disclosure, an electronic device is provided, comprising: a processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is in operation, the processor communicates with the memory via the bus, and when the machine-readable instructions are executed by the processor, the steps in the above-described remaining useful life prediction method are performed.

[0046] According to another aspect of this disclosure, a computer-readable storage medium is provided that stores a computer program, which, when executed by a processor, performs the steps in the above-described method for predicting remaining useful life.

[0047] According to another aspect of this disclosure, a computer program product is provided, which is stored in a storage medium and executed by at least one processor to implement the steps in the above-described method for predicting remaining useful life.

[0048] As will be described in detail below, there are methods, apparatus, devices, media, and products for predicting remaining useful life according to embodiments of this disclosure. In embodiments of this disclosure, sensor data can be pulse-coded to obtain a pulse sequence. This pulse sequence describes the dynamic changes in the operating state of industrial equipment at different points in time. The pulse sequence is then input to a spiking neuron to determine the state parameters of the spiking neuron, thereby allowing the prediction of the remaining useful life of the industrial equipment based on these state parameters. By pulse-coding sensor data, the temporal dependencies of the operating state can be effectively learned, and the remaining useful life of the industrial equipment can be predicted based on the state parameters of the spiking neuron. This allows for the extraction of fine-grained temporal features from the pulse sequence, thereby improving the prediction accuracy and precision of the remaining useful life.

[0049] It should be understood that both the foregoing general description and the following detailed description are exemplary and intended to provide further illustration of the claimed technology. Attached Figure Description

[0050] The above and other objects, features, and advantages of this disclosure will become more apparent from the more detailed description of the embodiments thereof in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this disclosure and form part of the specification. They are used together with the embodiments of this disclosure to explain the disclosure and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0051] Figure 1 This is a flowchart illustrating a method for predicting remaining useful life according to an embodiment of the present disclosure.

[0052] Figure 2 This is a further illustration of the architecture of the Recurrent Spiking Neural Network (RSNN) in the remaining useful life prediction method of this disclosure embodiment.

[0053] Figure 3 This is a flowchart illustrating the remaining useful life prediction process in the remaining useful life prediction method of the present disclosure embodiments.

[0054] Figure 4 This is a block diagram illustrating a remaining useful life prediction device according to an embodiment of the present disclosure.

[0055] Figure 5 This is a hardware block diagram illustrating an electronic device according to an embodiment of the present disclosure.

[0056] Figure 6 This is a schematic diagram illustrating a computer program product according to an embodiment of the present disclosure. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of this disclosure more apparent, exemplary embodiments according to this disclosure will now be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this disclosure, and not all embodiments of this disclosure. It should be understood that this disclosure is not limited to the exemplary embodiments described herein.

[0058] To facilitate understanding of this embodiment, a detailed description of the remaining useful life prediction method disclosed in this disclosure embodiment will be provided first. The execution subject of the remaining useful life prediction method provided in this disclosure embodiment is generally an electronic device with a certain computing power, such as a terminal device, a server, or other processing device. In some possible implementations, the remaining useful life prediction method can be implemented by a processor calling computer-readable instructions stored in memory.

[0059] See Figure 1The diagram shows a flowchart of a method for predicting remaining useful life provided in an embodiment of this disclosure. The method includes steps S101 to S104, wherein:

[0060] Step S101: Acquire sensor data from industrial equipment; wherein the sensor data is used to indicate the operating status of the industrial equipment.

[0061] Here, industrial equipment includes aircraft engines, gas turbines, and key units in manufacturing production lines, which are not limited in this disclosure.

[0062] The industrial equipment is equipped with multiple sensors, such as pressure sensors, temperature sensors, and speed sensors.

[0063] Accordingly, sensor data refers to data collected from sensors, such as pressure values ​​output by a pressure sensor and temperature values ​​output by a temperature sensor. Furthermore, the sensor data in this disclosure includes different combinations of operating conditions and failure modes to cover the complete life cycle of industrial equipment.

[0064] The embodiments disclosed herein can preprocess the sensor data, for example, through normalization, redundancy removal, and sliding window segmentation. Each sensor may correspond to one or more data transmission channels, and correspondingly, sensor data may also correspond to one or more data transmission channels.

[0065] The following preprocessing procedures can be used in this disclosure:

[0066] The variance of each data transmission channel or sensor data over its entire lifecycle is statistically analyzed, and channels or sensor data that remain unchanged are removed. Extreme value normalization is applied to sensor data corresponding to one or more sensors within the same application scenario to mitigate distribution drift caused by transitions between different application scenarios. To avoid excessively large predicted RUL values ​​in the early stages, which could lead to gradient dilution, and to address the uncertainty of the degradation time of industrial equipment, a maximum RUL threshold can be set in the early stages of the industrial equipment's lifecycle to limit the predicted RUL value from being too large. For example, the RUL threshold can be set to 125. A sliding window with a window length T = 30 and a time step s = 1 is used to segment the sensor data, and the data from the last time window is selected for prediction. For example, if there are 120 time steps of data, the data from the first 90 time steps can be used to predict the RUL of the 91st time step.

[0067] In addition, the implementer of the remaining useful life prediction method in this disclosure can be industrial equipment or third-party equipment, which integrates a neural network model, such as a recurrent spiking neural network (RSNN).

[0068] In this embodiment of the disclosure, industrial equipment can use sensor data collected by sensors as the data basis and preprocess the sensor data to ensure the continuity and real-time nature of the sensor data degradation trend.

[0069] Step S102: Perform pulse encoding on the sensor data to obtain a pulse sequence; wherein, the pulse sequence is used to describe the dynamic changes of the operating state at different time points.

[0070] Here, each pulse in the pulse sequence represents the operating status of industrial equipment at a specified point in time or during a specified time period, so as to describe the dynamic changes in the operating status in the form of pulses.

[0071] Since the sensor data originates from multiple sensors, the sensor data corresponding to each sensor can be pulse-coded to obtain multiple pulse sequences. Alternatively, the sensor data corresponding to each sensor can be integrated, and the integrated result can be pulse-coded to obtain a pulse sequence. This disclosure does not require this.

[0072] This disclosure does not limit the pulse coding method; for example, timing coding, frequency coding, or delay coding will be given as an example later.

[0073] In this embodiment of the disclosure, industrial equipment can convert continuous sensor data into discrete pulse events in the manner described above, so as to maintain the sparsity of pulse events and thus avoid redundant calculations.

[0074] Step S103: Input the pulse sequence into the spiking neuron to determine the state parameters of the spiking neuron.

[0075] Here, state parameters are used to describe the activity state of spiking neurons, such as membrane potential and total number of pulse releases, which will be described further later.

[0076] As described above, the neural network model used in this disclosure is RSNN. The spiking neurons in RSNN are arranged in layers. The original RSNN architecture includes an input encoding layer, a hidden layer, and an output decoding layer. There may be one or more hidden layers between the input encoding layer and the output decoding layer.

[0077] For the hidden layer, this disclosure employs a sparsely cyclically connected Leaky Integrate and Fire (LIF) neuron model to construct the liquid layer, which can be understood as an LSM (Liquid State Machine) architecture. The LIF neuron model is a biological neuron-based model used to simulate the dynamic behavior of neurons. This allows for the dynamic representation of the operating state of industrial equipment by storing and amplifying the temporal differences in pulse sequences through the dynamics of the LSM.

[0078] In this embodiment of the disclosure, industrial equipment can convert pulse sequences into state parameters of spiking neurons, thereby achieving natural encoding of temporal information and thus better capturing and processing dynamic features in pulse sequences.

[0079] Step S104: Based on the state parameters, predict the remaining service life of the industrial equipment.

[0080] In this embodiment of the disclosure, industrial equipment can use the state parameters of spiking neurons to predict the length of time from the current point in time until it fails or requires major maintenance, i.e., the remaining service life, in order to reduce the downtime of industrial equipment and reduce the maintenance cost of industrial equipment.

[0081] In the above embodiments, pulse coding can be performed on sensor data to obtain a pulse sequence, which can be used to characterize the temporal dependence of the operating state. The pulse sequence is then input into a spiking neuron, which can predict the remaining service life of industrial equipment based on the state parameters of the spiking neuron. This allows for the extraction of fine-grained temporal features from the pulse sequence, thereby improving the prediction accuracy and precision of the remaining service life.

[0082] In an optional implementation, the above steps input the pulse sequence to the spiking neuron to determine the state parameters of the spiking neuron, specifically including the following steps:

[0083] The pulse sequence is mapped to a target vector space; wherein the target vector space is used to describe the membrane potential of the spiking neuron at different time points;

[0084] The total number of pulse releases by the spiking neuron is determined based on the potential changes of the membrane potential.

[0085] The membrane potential and the total number of pulse releases are used as the state parameters.

[0086] As mentioned above, this disclosure introduces LSM into the RSNN architecture. Based on the dynamic idea of ​​LSM, this disclosure first projects the pulse sequence to the input synapse of the liquid layer, which is composed of M heterogeneous LIF neurons (usually much larger than the input dimension N). This can be understood as the spiking neurons of the liquid layer, and they form a dynamic system through random and sparse recursive connections.

[0087] The input dimension can be understood as the number of spiking neurons in the input coding layer. Each spiking neuron in the input coding layer can correspond to one or more pulse sequences. Correspondingly, each spiking neuron in the input coding layer can also correspond to one or more spiking neurons in the liquid layer. The correspondence between the pulse sequences, the spiking neurons in the input coding layer, and the spiking neurons in the liquid layer can be adjusted as needed.

[0088] When the pulse sequence enters the spiking neurons of the liquid layer, the membrane potential of each spiking neuron in the liquid layer undergoes transient fluctuations under the combined effects of leakage current, pulse sequence, and cyclic feedback, thus forming a liquid morphology that evolves over time.

[0089] Formally, if the pulse sequence is u(t), then the cyclically connected liquid layers will form a nonlinear mapping from low dimension to high dimension, that is, a mapping process from the input dimension of N dimensions to the order of M dimensions: x M (t)=L M (u(t)), where L M Let x be the mapping function that maps a pulse sequence. M (t) is the high-dimensional liquid state vector of the liquid layer at time t. This high-dimensional liquid state vector can essentially be understood as the membrane potential of the spiking neurons of the liquid layer at time t. This will be described in detail later and will not be explained further here.

[0090] The high-dimensional liquid state vectors at different time points can form a vector space, which is the target vector space. This allows the pulse sequence to be mapped to a significantly separated trajectory. At the same time, the natural decay of the membrane potential in the target vector space can ensure that historical information gradually dissipates. This historical information is used to describe the dynamic changes of the operating state at historical moments, thereby achieving decay memory rather than static storage.

[0091] This disclosure addresses the potential changes of membrane potential, which can be described using differential equations to balance biological interpretability and computational simplicity: Where, τ m V(t) represents the membrane time constant controlling the membrane potential decay rate, and V(t) is the membrane potential at time t. rest R represents the reset potential of the spiking neurons in the liquid layer. m Let I(t) be the membrane resistance, and let I(t) be the synaptic current, i.e., the pulse sequence.

[0092] The above formula is used to continuously accumulate the membrane potential by integrating the synaptic current. When the membrane potential increases due to the pulse sequence or recursive feedback and reaches the preset potential threshold, the spiking neuron will immediately release a pulse and reset the membrane potential to the reference level, i.e., reset the potential. At the same time, it enters the natural decay stage determined by the membrane time constant. Based on the continuous cycle of the above process, the number of pulse releases is accumulated to obtain the total number of pulse releases of the spiking neuron.

[0093] Because the aforementioned computational processes occur during pulse triggering and potential updates, spiking neurons possess event-driven sparse computational characteristics, which can significantly reduce the number of operations and computational resources. Moreover, the pulse sequence carries dynamic characteristics, enabling the capture of fine-grained changes reflected by temperature, pressure, or speed sensors during the degradation process of industrial equipment. This lays a high-fidelity temporal input foundation for the mapping process of the pulse sequence to the target vector space, further providing neurodynamic support for accurately characterizing degradation patterns.

[0094] In the above embodiments, by leveraging the dynamics of LSM, low-dimensional pulse sequences are nonlinearly mapped to high-dimensional liquid state vectors containing fading memories, thereby fully explaining the degradation process and long- and short-term temporal dependencies of industrial equipment. Furthermore, this disclosure selects membrane potential and the total number of pulse releases as state parameters of the spiking neuron, taking into account both the macroscopic temporal characteristics carried by the discrete pulse sequence and the microscopic subthreshold changes reflected by the membrane potential, achieving high-precision and stable lifetime prediction.

[0095] In an optional implementation, the above steps map the pulse sequence to the target vector space, specifically including the following steps:

[0096] Obtain the pulse release state of multiple spiking neurons at the first time point;

[0097] The pulse release state and the target pulse sequence are weighted and summed to obtain the target weighted result; wherein, the target pulse sequence is the pulse sequence at the first time point;

[0098] Determine the membrane potential decay rate of the spiking neuron at the first time point;

[0099] The target weighted result and the membrane potential decay rate are summed to obtain the membrane potential of the spiking neuron at the second time point; wherein the second time point is after the first time point.

[0100] The mapping process for pulse sequences can be calculated using the following formula:

[0101] Among them, s j(t) represents the pulse release state of the j-th spiking neuron in the liquid layer at the first time point t. When this spiking neuron releases a pulse, s j When (t) = 1, s is the pulse neuron that has not released a pulse. j (t) = 0; x k (t) represents the pulse sequence received by the k-th spiking neuron in the input coding layer at the first time point t. The target pulse sequence is obtained by integrating the pulse sequences received by all spiking neurons in the input coding layer at time t. The weights for the recursive connections between spiking neurons within the liquid layer are the first weights mentioned later. This is used to describe the degree of influence of the target pulse sequence on the spiking neurons, i.e., the second weight mentioned later; M is the number of spiking neurons in the liquid layer, and N is the number of spiking neurons in the input coding layer; Weighted results for the objective; V i (t) represents the membrane potential of the i-th spiking neuron in the liquid layer at time t; Δt is the time difference between the first time point and the second time point t+Δt; τ m The membrane time constant represents the rate at which the membrane potential decays. V is the membrane potential decay rate; i (t+Δt) represents the membrane potential of the i-th spiking neuron in the liquid layer at the second time point t+Δt. The above processing procedure will be further described later.

[0102] In the above embodiments, it can be ensured that the potential that has not triggered the preset potential threshold decreases exponentially over time, enabling the spiking neurons to have the function of memory-forgetting balance, and retaining short-term historical information without excessive accumulation. Moreover, through the cyclical interaction between spiking neurons, the nonlinear integration of historical input data can be promoted, thereby achieving the linear separability of membrane potentials that change over time, and improving the accuracy of remaining lifetime prediction.

[0103] In an optional implementation, the above steps involve weighted summation of the pulse release state and the target pulse sequence to obtain a target weighted result, specifically including the following steps:

[0104] The first weight and the corresponding pulse release state are weighted to obtain a first weighted result; wherein, the first weight is used to indicate the connection strength between the spiking neuron and other spiking neurons;

[0105] The second weight and the corresponding target pulse sequence are weighted to obtain a second weighted result; wherein, the second weight is used to indicate the degree of influence of the target pulse sequence on the spiking neuron;

[0106] The first weighted result and the second weighted result are summed to obtain the target weighted result.

[0107] Here, the first weighted result is Among them, s j (t) represents the pulse release state of the j-th spiking neuron in the liquid layer at the first time point t, and M represents the number of spiking neurons in the liquid layer. It is the first weight.

[0108] For the second time point, the pulse release state of each spiking neuron in the liquid layer and its corresponding first weight can be weighted, and the weighted results can be integrated together to obtain the first weighted result.

[0109] Here, the second weighted result is Where, x k (t) represents the pulse sequence received by the k-th spiking neuron in the input coding layer at the first time point t, where N is the number of spiking neurons in the input coding layer. It is the second weight.

[0110] For the second time point, the pulse sequence received by each spiking neuron in the input coding layer and its corresponding second weight can be weighted, and the weighted results can be integrated together to obtain the second weighted result.

[0111] In this embodiment of the disclosure, the pulse release state and the first weight are weighted to obtain a first weighted result, and the target pulse sequence and the second weight are weighted to obtain a second weighted result, so as to obtain the target weighted result. This can comprehensively consider the activity state of other spiking neurons and the pulse sequence of other spiking neurons, so that these two aspects can be divided into the membrane potential update process to improve the accuracy of membrane potential update.

[0112] In an optional implementation, the above steps for determining the membrane potential decay rate of the spiking neuron at the first time point specifically include the following steps:

[0113] Determine the time difference between the first time point and the second time point;

[0114] The time difference and membrane time constant are subjected to an exponential operation to obtain the calculation result;

[0115] The product of the calculation result and the membrane potential of the spiking neuron at the first time point is determined as the membrane potential decay rate.

[0116] This disclosure describes the process for calculating the membrane potential decay rate, namely Where Δt is the time difference between the first and second time points, Vi (t) represents the membrane potential of the i-th spiking neuron in the liquid layer at time t, τ m The membrane time constant is used to represent the rate of membrane potential decay. This is the result of the calculation.

[0117] In this embodiment of the disclosure, the membrane potential decay rate is determined based on the time difference, membrane time constant, and membrane potential, and the membrane potential decay rate is incorporated into the membrane potential update process, thereby further improving the accuracy of membrane potential update.

[0118] In the above implementation, the membrane potential at the current moment is updated by combining the membrane potential at historical moments, the pulse release state, and the input pulse sequence. While maintaining the sparsity of pulse events, historical inputs are continuously fused with cyclic weighting coefficients, so that key information is amplified and separated in the target vector space.

[0119] In an optional implementation, the above steps perform pulse encoding on the sensor data to obtain a pulse sequence, specifically including the following steps:

[0120] Extract operational status indicators from the sensor data according to at least one extraction dimension;

[0121] Map the preprocessing results of the operating status indicators to the pulse transmission time;

[0122] The pulse sequence is generated based on the pulse transmission time.

[0123] Here, the extraction dimensions can be adjusted according to the actual application scenario. Specifically, the extraction dimensions may include the magnitude of data changes, statistical features, or keywords that can reflect the key performance of industrial equipment. Among them, statistical features can be the mean, maximum and minimum values, and variance.

[0124] In this embodiment of the disclosure, the sensor data can be preprocessed first, and then the process of extracting the operating status indicators from the sensor data can be executed, so that the preprocessed results of the operating status indicators can be obtained directly.

[0125] In addition, operational status indicators can be extracted from sensor data first, and then preprocessed to obtain preprocessed operational status indicator results. The preprocessing process has been mentioned earlier in this disclosure and will not be described further here. Furthermore, this disclosure does not limit the method of obtaining the preprocessed results.

[0126] In this embodiment of the disclosure, a delayed coding method is used to extract the operating status index from the sensor data and map the operating status index to the pulse dimension to generate an encoded pulse sequence based on the pulse transmission time. This allows the pulse sequence to carry more accurate dynamic features, making it easier to make high-precision predictions based on the pulse sequence.

[0127] In an optional implementation, the above steps map the preprocessing results of the operating status indicators to the pulse transmission time, specifically including the following steps:

[0128] The pulse transmission time is determined based on the preprocessing result and the mapping time; wherein, the mapping time is the time from the preprocessing result to the pulse transmission time.

[0129] in, For the preprocessing result corresponding to the i-th spiking neuron in the input coding layer, t i Let T be the pulse transmission time of the i-th neuron in the input coding layer. enc The mapping time can also be the total duration of the time window mentioned in the preprocessing process above.

[0130] In the case where a sensor corresponds to a neuron in the input coding layer This can be the preprocessing result of the sensor data from the i-th sensor, and the preprocessing result can be 0 to 1. The higher the preprocessing result, the earlier the trigger pulse, that is, the earlier the pulse transmission time. This allows the pulse transmission time to accurately carry the operating status indicators, while ensuring that only a very small number of neurons in the entire input coding layer are active at any given time.

[0131] Furthermore, for situations where the dynamic range of operational status indicators or preprocessing results is large, a group-based delayed coding approach can be employed. This involves driving several threshold-increasing spiking neurons in parallel with the same operational status indicator or preprocessing result, causing the pulse density to increase non-linearly with the finer granularity of the numerical values, thereby improving resolution while maintaining sparsity. This delayed coding method automatically embeds the time dimension, preserving the rhythmic characteristics of the degradation curve completely without additional time embedding or positional coding, and preventing other data (excluding operational status indicators) from generating pulses, significantly reducing subsequent computational load.

[0132] In an optional implementation, the above steps, based on the state parameters, predict the remaining service life of the industrial equipment, specifically including the following steps:

[0133] The membrane potential and total number of pulse releases of the spiking neurons are weighted and summed to obtain a third weighted result;

[0134] The remaining useful life is predicted based on the third weighted result.

[0135] This disclosure employs a hybrid decoding strategy (membrane potential - total number of pulse releases) in the output decoding layer to obtain coherent and high-accuracy lifetime estimation. During the prediction process, the spiking neurons in the output decoding layer accumulate their own pulse release counts to obtain the total number of pulse releases, reflecting the macroscopic temporal activity pattern; on the other hand, they read the instantaneous membrane potential of the spiking neurons to capture subtle twitches at the subthreshold level.

[0136] In this embodiment of the disclosure, for the two aspects mentioned above, linear fusion is performed according to the weight coefficient (0.3-0.6), and then a third weighted result is obtained through a single fully connected layer, and the third weighted result is mapped to the RUL result.

[0137] In the above implementation, the total number of pulse releases ensures that the output changes monotonically with respect to the overall trend of the historical pulse stream, while the membrane potential compensates for the quantization errors and jitter that are prone to occur when relying solely on the total number of pulse releases, achieving smooth and continuous prediction driven by sparse event streams. Simultaneously, this hybrid approach, through the continuous transmission of the membrane potential gradient during backpropagation, contributes to stable training. The entire decoding process is easily implemented on FPGAs (Field Programmable Gate Arrays) or neuromorphic chips, and can guarantee real-time performance and energy efficiency in resource-constrained scenarios, such as millisecond-level low latency and microwatt-level power consumption.

[0138] In an optional implementation, the above steps further include the following steps:

[0139] Differentiate the pulse release state function to obtain the derivative result;

[0140] Based on the product of the derivative result and the learning rate, the first weight and the second weight are optimized to obtain the optimized first weight and the optimized second weight; wherein, the learning rate is used to indicate the adjustment range of the first weight and the second weight;

[0141] The optimized first weight is used as the weight of the pulse release state, and the optimized second weight is used as the weight of the target pulse sequence.

[0142] Since the pulse release state function of a spiking neuron is a step function and is inherently non-differentiable, traditional gradient descent strategies cannot be directly applied. Therefore, this disclosure introduces an alternative gradient approach in the backpropagation through time (BPTT) process: forward propagation still uses the actual LIF release mechanism, while the derivative of the step function is replaced by a differentiable smooth curve during backpropagation, thus achieving error chain backpropagation while maintaining accurate pulse timing.

[0143] Specifically, the following formula can be used as an approximation during the backpropagation phase:

[0144] Where S(V) is the pulse release state function, γ is an adjustable hyperparameter controlling the slope and smoothness of the substitution gradient, and V is the membrane potential. threshold The preset potential threshold is used. This smooth approximation avoids the zero gradient problem and suppresses excessive gradient oscillations at the pulse edges, making the gradient design calculation stable and reliable.

[0145] Based on this, the product of the learning rate and the derivative can be determined, and the first weight and the second weight can be subtracted from the product to obtain the optimized first weight and the optimized second weight.

[0146] After adopting the above-mentioned alternative gradient, the weight coefficients can be iteratively updated using standard BPTT. The loss function is mainly based on the mean square error (MSE) between the predicted and actual RUL values, and an asymmetric penalty term can be optionally added to emphasize the punishment of overly optimistic predictions.

[0147] The basic form is: Where B is the batch size, i.e., the number of data samples processed simultaneously during one training cycle, and y b and Let represent the true RUL value and the predicted RUL value of the b-th data sample, respectively.

[0148] During training, the Adam optimizer can be used in conjunction with gradient norm pruning to prevent gradient explosion. At the same time, sparse regularization constraints are applied to the average output firing rate, which further reduces training energy consumption while ensuring expressive power and provides a reliable parameter basis for subsequent low-power real-time inference.

[0149] Based on this, this disclosure comprehensively considers two dimensions—prediction accuracy and computational economy—when evaluating the merits of the model. Regarding prediction accuracy, the root mean square error (RMSE) is used as the overall error measure. This indicator is the square root of the mean of the squared deviations, and therefore more sensitive to large errors.

[0150] The specific formula is as follows: At the same time, an asymmetric penalty function is introduced, specifically expressed mathematically as follows: Where B is the batch size, i.e., the number of data samples processed simultaneously during one training cycle, and y b and Let represent the true RUL value and the predicted RUL value of the b-th data sample, respectively.

[0151] This penalty function imposes a more severe penalty on overly optimistic cases (i.e., overestimating the RUL), aligning with maintenance principles in industrial safety scenarios. Regarding computational efficiency, this disclosure calculates the floating-point operations required for model inference to measure theoretical computational complexity; it also tracks memory usage to reflect the hardware storage resource requirements during deployment; and it records single-sample inference latency to evaluate the response speed of real-time applications. By simultaneously examining these performance and efficiency metrics, the applicability of each algorithm in practical application scenarios can be comprehensively revealed.

[0152] Considering both accuracy and efficiency, RSNN can significantly reduce computation and storage overhead while ensuring higher prediction accuracy, providing a practical solution for online device health management in resource-constrained scenarios.

[0153] Reference Figure 2 The diagram shown is an architecture diagram of the RSNN in the remaining useful life prediction process provided in this embodiment of the present disclosure. The architecture diagram includes an input encoding layer, a hidden layer (liquid layer) and an output decoding layer. The processing procedures of each layer in the architecture diagram can be referred to above, and will not be described further here.

[0154] The following is combined with Figure 3 The process for predicting the remaining useful life described above is as follows:

[0155] S301: Acquire sensor data from industrial equipment.

[0156] Sensor data is used to indicate the operating status of industrial equipment.

[0157] S302: Pulse encoding is performed on the sensor data to obtain a pulse sequence.

[0158] Among them, the pulse sequence is used to describe the dynamic changes in the operating state at different points in time.

[0159] S303: Maps the pulse sequence to the target vector space.

[0160] The target vector space is used to describe the membrane potential of spiking neurons at different time points.

[0161] S304: Determine the total number of pulse releases of a spiking neuron based on the potential changes in membrane potential.

[0162] S305: Use membrane potential and total number of pulse releases as state parameters.

[0163] S306: Predict the remaining service life of industrial equipment based on condition parameters.

[0164] As can be seen from the above description, the technical solution disclosed herein has the following advantages:

[0165] (1) High prediction accuracy: The combination of high-dimensional liquid state and hybrid decoding fully explores the fine-grained temporal features of degraded sequences, and the prediction accuracy is significantly higher than that of existing deep learning and traditional machine learning models.

[0166] (2) Excellent real-time performance and energy efficiency: Event-driven and sparse pulse computing significantly reduce floating-point operations and storage requirements; millisecond-level inference can be achieved on graphics processing units (GPUs), FPGAs or neuromorphic chips to meet the needs of online monitoring and edge deployment.

[0167] (3) Strong model generalization and adaptability: The liquid state machine structure does not require reconstruction of the neural network for specific equipment. It can be migrated to different working conditions or new equipment by adjusting only a few hyperparameters, reducing development and maintenance costs.

[0168] (4) Stable training and easy implementation: The alternative gradient method is used to solve the problem of nondifferentiability of spiking networks. Combined with the end-to-end training process, it not only ensures convergence stability, but also avoids complex manual design and feature engineering.

[0169] (5) Interpretability and visualization friendliness: Hybrid decoding preserves continuous information of membrane potential, and combined with liquid layer pulse mode, it can intuitively present the dynamic evolution of degradation process, which is convenient for operation and maintenance personnel to understand and make decisions.

[0170] (6) Adapt to low-power hardware: The sparse network and event-driven computing characteristics are compatible with neuromorphic chips and other low-power processors, enabling long-term operation of the device without frequent maintenance or power supply replacement.

[0171] Based on the same inventive concept, this disclosure also provides a remaining useful life prediction device corresponding to the remaining useful life prediction method. Since the principle of the device in this disclosure is similar to the remaining useful life prediction method described above, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be described again.

[0172] Reference Figure 4 The diagram shown is a schematic representation of a remaining useful life prediction device provided in an embodiment of this disclosure. The device includes: an acquisition unit 40, a pulse coding unit 41, a parameter determination unit 42, and a prediction unit 43; wherein:

[0173] An acquisition unit is used to acquire sensor data from industrial equipment; wherein the sensor data is used to indicate the operating status of the industrial equipment.

[0174] A pulse coding unit is used to perform pulse coding on the sensor data to obtain a pulse sequence; wherein, the pulse sequence is used to describe the dynamic changes of the operating state at different time points;

[0175] A parameter determination unit is used to input the pulse sequence into the spiking neuron in order to determine the state parameters of the spiking neuron;

[0176] A prediction unit is used to predict the remaining service life of the industrial equipment based on the state parameters.

[0177] In one possible implementation, the device is also used for:

[0178] The pulse sequence is mapped to a target vector space; wherein the target vector space is used to describe the membrane potential of the spiking neuron at different time points;

[0179] The total number of pulse releases by the spiking neuron is determined based on the potential changes of the membrane potential.

[0180] The membrane potential and the total number of pulse releases are used as the state parameters.

[0181] In one possible implementation, the device is also used for:

[0182] Obtain the pulse release state of multiple spiking neurons at the first time point;

[0183] The pulse release state and the target pulse sequence are weighted and summed to obtain the target weighted result; wherein, the target pulse sequence is the pulse sequence at the first time point;

[0184] Determine the membrane potential decay rate of the spiking neuron at the first time point;

[0185] The target weighted result and the membrane potential decay rate are summed to obtain the membrane potential of the spiking neuron at the second time point; wherein the second time point is after the first time point.

[0186] In one possible implementation, the device is also used for:

[0187] The first weight and the corresponding pulse release state are weighted to obtain a first weighted result; wherein, the first weight is used to indicate the connection strength between the spiking neuron and other spiking neurons;

[0188] The second weight and the corresponding target pulse sequence are weighted to obtain a second weighted result; wherein, the second weight is used to indicate the degree of influence of the target pulse sequence on the spiking neuron;

[0189] The first weighted result and the second weighted result are summed to obtain the target weighted result.

[0190] In one possible implementation, the device is also used for:

[0191] Determine the time difference between the first time point and the second time point;

[0192] The time difference and membrane time constant are subjected to an exponential operation to obtain the calculation result;

[0193] The product of the calculation result and the membrane potential of the spiking neuron at the first time point is determined as the membrane potential decay rate.

[0194] In one possible implementation, the device is also used for:

[0195] Extract operational status indicators from the sensor data according to at least one extraction dimension;

[0196] Map the preprocessing results of the operating status indicators to the pulse transmission time;

[0197] The pulse sequence is generated based on the pulse transmission time.

[0198] In one possible implementation, the device is also used for:

[0199] The pulse transmission time is determined based on the preprocessing result and the mapping time; wherein, the mapping time is the time from the preprocessing result to the pulse transmission time.

[0200] In one possible implementation, the device is also used for:

[0201] The membrane potential and total number of pulse releases of the spiking neurons are weighted and summed to obtain a third weighted result;

[0202] The remaining useful life is predicted based on the third weighted result.

[0203] In one possible implementation, the device is also used for:

[0204] Differentiate the pulse release state function to obtain the derivative result;

[0205] Based on the product of the derivative result and the learning rate, the first weight and the second weight are optimized to obtain the optimized first weight and the optimized second weight; wherein, the learning rate is used to indicate the adjustment range of the first weight and the second weight;

[0206] The optimized first weight is used as the weight of the pulse release state, and the optimized second weight is used as the weight of the target pulse sequence.

[0207] The processing flow of each module in the device and the interaction flow between each module can be referred to the relevant descriptions in the above method embodiments, and will not be detailed here.

[0208] Corresponding to Figure 1 In addition to the remaining useful life prediction method, this disclosure also provides an electronic device 50, such as... Figure 5 The diagram shown is a structural schematic of the electronic device 50 provided in an embodiment of this disclosure, including:

[0209] The system includes a processor 51, a memory 52, and a bus 53. The memory 52 stores execution instructions and includes main memory 521 and external memory 522. The main memory 521, also called internal memory, temporarily stores the computational data in the processor 51, as well as data exchanged with external memory such as a hard disk. The processor 51 exchanges data with the external memory 522 through the main memory 521. When the electronic device 50 is running, the processor 51 communicates with the memory 52 through the bus 53, causing the processor 51 to execute the following instructions:

[0210] Acquire sensor data from industrial equipment; wherein the sensor data is used to indicate the operating status of the industrial equipment;

[0211] The sensor data is pulse encoded to obtain a pulse sequence; the pulse sequence is used to describe the dynamic changes of the operating state at different points in time.

[0212] The pulse sequence is input into the spiking neuron to determine the state parameters of the spiking neuron;

[0213] Based on the state parameters, the remaining service life of the industrial equipment is predicted.

[0214] This disclosure also provides a computer-readable storage medium storing a computer program that, when executed by a processor, performs the steps of the remaining useful life prediction method described in the above-described method embodiments. The storage medium may be a volatile or non-volatile computer-readable storage medium.

[0215] This disclosure also provides a computer program product 60, such as... Figure 6 The diagram shown is a schematic diagram of the structure of a computer program product 60 provided in an embodiment of this disclosure. The computer program product 60 carries a computer program 61. The program included in the computer program 61 can be used to execute the steps of the remaining useful life prediction method described in the above method embodiments. For details, please refer to the above method embodiments, which will not be repeated here.

[0216] The remaining useful life prediction method, apparatus, device, medium, and product according to embodiments of the present disclosure have been described above with reference to the accompanying drawings. First, sensor data from industrial equipment is acquired; wherein the sensor data indicates the operating state of the industrial equipment; then, the sensor data is pulse-coded to obtain a pulse sequence; wherein the pulse sequence describes the dynamic changes of the operating state at different time points; and the pulse sequence is input to a spiking neuron to determine the state parameters of the spiking neuron; finally, based on the state parameters, the remaining useful life of the industrial equipment is predicted. By pulse-coding the sensor data, the temporal dependencies of the operating state can be effectively learned, and the remaining useful life of the industrial equipment can be predicted based on the state parameters of the spiking neuron, thereby mining fine-grained temporal features in the pulse sequence and improving the prediction accuracy and precision of the remaining useful life.

[0217] The basic principles of this disclosure have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this disclosure are merely examples and not limitations, and should not be considered as essential features of each embodiment of this disclosure. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the scope of this disclosure to the necessity of employing the aforementioned specific details for implementation.

[0218] The block diagrams of devices, apparatuses, devices, and systems disclosed herein are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.

[0219] Additionally, as used herein, the “or” used in a list of items beginning with “at least one” indicates a separate list, such that a list of, for example, “at least one of A, B, or C” means A or B or C, or AB or AC or BC, or ABC (i.e., A and B and C). Furthermore, the word “exemplary” does not imply that the described example is preferred or better than other examples.

[0220] It should also be noted that in the systems and methods of this disclosure, the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered as equivalent solutions to this disclosure.

[0221] Various changes, substitutions, and modifications can be made to the technology described herein without departing from the teachings defined by the appended claims. Furthermore, the scope of the claims of this disclosure is not limited to the specific aspects of the processes, machines, manufactures, events, means, methods, and actions described above. Currently existing or later-developed processes, machines, manufactures, events, means, methods, or actions that perform substantially the same function or achieve substantially the same result as the corresponding aspects described herein can be utilized. Therefore, the appended claims include such processes, machines, manufactures, events, means, methods, or actions within their scope.

[0222] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of this disclosure. Therefore, this disclosure is not intended to be limited to the aspects shown herein, but rather to be carried out within the widest scope consistent with the principles and novel features disclosed herein.

[0223] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this disclosure to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations therein.

Claims

1. A method for predicting remaining useful life, characterized in that, include: Acquire sensor data from industrial equipment; wherein the sensor data is used to indicate the operating status of the industrial equipment; The sensor data is pulse encoded to obtain a pulse sequence; the pulse sequence is used to describe the dynamic changes of the operating state at different points in time. The pulse sequence is mapped to a target vector space; wherein the target vector space is used to describe the membrane potential of the spiking neuron at different time points; The total number of pulse releases by the spiking neuron is determined based on the potential changes of the membrane potential. The membrane potential and the total number of pulse releases are used as state parameters of the spiking neuron; Based on the state parameters, the remaining service life of the industrial equipment is predicted; The step of mapping the pulse sequence to the target vector space includes: Obtain the pulse release state of multiple spiking neurons at the first time point; The pulse release state and the target pulse sequence are weighted and summed to obtain the target weighted result; wherein, the target pulse sequence is the pulse sequence at the first time point; Determine the membrane potential decay rate of the spiking neuron at the first time point; The target weighted result and the membrane potential decay rate are summed to obtain the membrane potential of the spiking neuron at the second time point; wherein the second time point is after the first time point.

2. The method according to claim 1, characterized in that, The weighted summation of the pulse release state and the target pulse sequence to obtain the target weighted result includes: The first weight and the corresponding pulse release state are weighted to obtain a first weighted result; wherein, the first weight is used to indicate the connection strength between the spiking neuron and other spiking neurons; The second weight and the corresponding target pulse sequence are weighted to obtain a second weighted result; wherein, the second weight is used to indicate the degree of influence of the target pulse sequence on the spiking neuron; The first weighted result and the second weighted result are summed to obtain the target weighted result.

3. The method according to claim 1, characterized in that, Determining the membrane potential decay rate of the spiking neuron at the first time point includes: Determine the time difference between the first time point and the second time point; The time difference and membrane time constant are subjected to an exponential operation to obtain the calculation result; The product of the calculation result and the membrane potential of the spiking neuron at the first time point is determined as the membrane potential decay rate.

4. The method according to claim 1, characterized in that, The step of pulse encoding the sensor data to obtain a pulse sequence includes: Extract operational status indicators from the sensor data according to at least one extraction dimension; Map the preprocessing results of the operating status indicators to the pulse transmission time; The pulse sequence is generated based on the pulse transmission time.

5. The method according to claim 4, characterized in that, The step of mapping the preprocessing results of the operating status indicators to the pulse transmission time includes: The pulse transmission time is determined based on the preprocessing result and the mapping time; wherein, the mapping time is the time from the preprocessing result to the pulse transmission time.

6. The method according to claim 1, characterized in that, The process of predicting the remaining service life of the industrial equipment based on the state parameters includes: The membrane potential and total number of pulse releases of the spiking neurons are weighted and summed to obtain a third weighted result; The remaining useful life is predicted based on the third weighted result.

7. The method according to claim 2, characterized in that, The method further includes: Differentiate the pulse release state function to obtain the derivative result; Based on the product of the derivative result and the learning rate, the first weight and the second weight are optimized to obtain the optimized first weight and the optimized second weight; wherein, the learning rate is used to indicate the adjustment range of the first weight and the second weight; The optimized first weight is used as the weight of the pulse release state, and the optimized second weight is used as the weight of the target pulse sequence.

8. A device for predicting remaining useful life, characterized in that, include: An acquisition unit is used to acquire sensor data from industrial equipment; wherein the sensor data is used to indicate the operating status of the industrial equipment. A pulse coding unit is used to perform pulse coding on the sensor data to obtain a pulse sequence; wherein, the pulse sequence is used to describe the dynamic changes of the operating state at different time points; A parameter determination unit is used to map the pulse sequence to a target vector space; wherein the target vector space is used to describe the membrane potential of the spiking neuron at different time points; the total number of pulse releases of the spiking neuron is determined based on the potential change of the membrane potential; and the membrane potential and the total number of pulse releases are used as state parameters of the spiking neuron. A prediction unit is configured to predict the remaining service life of the industrial equipment based on the state parameters; wherein, mapping the pulse sequence to a target vector space includes: Obtain the pulse release state of multiple spiking neurons at the first time point; The pulse release state and the target pulse sequence are weighted and summed to obtain the target weighted result; wherein, the target pulse sequence is the pulse sequence at the first time point; Determine the membrane potential decay rate of the spiking neuron at the first time point; The target weighted result and the membrane potential decay rate are summed to obtain the membrane potential of the spiking neuron at the second time point; wherein the second time point is after the first time point.

9. An electronic device, characterized in that, include: The device includes a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is in operation, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, the steps of the remaining useful life prediction method as described in any one of claims 1 to 7 are performed.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the remaining useful life prediction method as described in any one of claims 1 to 7.

11. A computer program product, characterized in that, The computer program product is stored in a storage medium and is executed by at least one processor to implement the remaining useful life prediction method as described in any one of claims 1 to 7.

Images