AI-based condition diagnosis system and method for vertical transportation equipment

Abstract

The present invention relates to a system and method for accurately diagnosing the condition of elevator equipment using artificial intelligence (AI). The present invention extracts feature variables from vibration or noise signal data sensed in elevator equipment in a normal state, calculates the Mahalanobis distance for the feature variables, and then, while training a learning model with the Mahalanobis distance data, infers the Mahalanobis distance data for vibration or noise signal data sensed in elevator equipment to be diagnosed from the learning model to diagnose whether the elevator equipment to be diagnosed is normal.

Description

Technology Field

[0001] The present invention relates to a system and method for diagnosing the condition of elevator equipment, and more particularly to a system and method for accurately diagnosing the condition of elevator equipment using artificial intelligence (AI). Background Technology

[0002] Lifting facilities such as elevators, escalators, and moving walkways are essential in modern buildings, enabling people and goods to move quickly and conveniently.

[0003] Accordingly, since a problem with the elevator equipment can cause serious harm to both human lives and cargo, the safety and condition diagnosis of the equipment are considered very important factors.

[0004] With the advancement of the Internet of Things (IoT) and artificial intelligence (AI) technologies, smart management technology that enables real-time diagnosis of elevator equipment conditions and predictive maintenance is gaining attention.

[0005] However, the current situation is limited to collecting sensor data from elevator equipment and comprehensively analyzing the collected data to diagnose faults.

[0006] Recently, technology for diagnosing the condition of elevator equipment based on mechanical analysis and diagnostic theories derived from the design specifications of elevator equipment has been proposed.

[0007] While such diagnostics can be expected to yield high-accuracy results due to the prior mechanical analysis, they are heavily influenced by mechanical characteristics, including the installation environment and lifespan of each elevator, and require the establishment of fixed rules that account for numerous variables, making it difficult to ensure system flexibility.

[0008] Furthermore, mechanical analysis of elevator equipment requires a sufficient amount of equipment failure and replacement history; however, since such history is often insufficient or missing, engineering-based diagnostic results sometimes fail to achieve a high level of accuracy.

[0009] Accordingly, the relevant technology field requires the development of technology that automatically analyzes the mechanical condition of elevator equipment by introducing artificial intelligence (AI) technology into the condition diagnosis of elevator equipment. Prior art literature

[0010] Registered Patent Publication No. 10-2437701 Registered Patent Publication No. 10-2651107 The problem to be solved

[0011] The present invention aims to provide an AI-based elevator equipment condition diagnosis system and method that diagnoses the condition of elevator equipment by analyzing vibration and noise data of components of elevator equipment using artificial intelligence (AI).

[0012] The present invention aims to provide an AI-based elevator equipment condition diagnosis system and method that utilizes AI technology to diagnose abnormalities in key components of elevator equipment in real time and predict the possibility of failure in advance.

[0013] The present invention aims to provide an AI-based elevator equipment condition diagnosis system and method that analyzes vibration and noise data collected during the operation of elevator equipment using an AI-based model to automatically classify normal and abnormal states and predict failure types.

[0014] The present invention aims to provide an AI-based elevator equipment condition diagnosis system and method that enables predictive maintenance for similar situations by learning the past failure history and patterns of elevator equipment using AI. means of solving the problem

[0015] An AI-based elevator facility condition diagnosis method according to an embodiment of the present invention comprises: sensing a vibration or noise signal in an elevator facility in a normal state; extracting a signal of a preset frequency band from the sensed signal; extracting feature parameters from the data of the extracted signal; calculating a Mahalanobis distance for the feature parameters; learning data of a Mahalanobis Space defined as a data set of the calculated Mahalanobis distances through a first learning model; additionally sensing a vibration or noise signal in an elevator facility to be diagnosed; extracting a signal of the frequency band from the additionally sensed signal; extracting feature parameters from the data of the extracted signal; calculating a Mahalanobis distance for the feature parameters; and determining whether the elevator facility to be diagnosed is in a normal state or an abnormal state by inferring the calculated Mahalanobis distance from the first learning model. The method includes the step of generating a predicted STFT image of future data based on current data of a signal additionally sensed from the elevator equipment under diagnosis using a second learning model when the elevator equipment under diagnosis is determined to be in an abnormal state; and the step of determining signs of failure of the elevator equipment under diagnosis through a third learning model based on the generated predicted STFT image.

[0016] In the present invention, the step of sensing a vibration or noise signal of the lifting equipment involves sensing a vibration signal or noise signal of a component at a sensor attached to a component constituting the lifting equipment.

[0017] In the present invention, the step of extracting the signal of the frequency band excludes signals unnecessary for the diagnosis of the elevator equipment and filters only the signals necessary for the diagnosis of the elevator equipment.

[0018] In the present invention, the step of extracting a signal of the frequency band comprises: a step of oversampling data at a sampling rate higher than a preset sampling rate; and a step of filtering the data of the frequency band signal from the oversampled data and downsampling it to the data of the preset sampling rate.

[0019] In the present invention, the first learning model includes a One-Class SVM (Support Vector Machine).

[0020] In the present invention, the One-Class SVM determines the elevator equipment to be diagnosed as being in a normal state if the Mahalanobis distance calculated in the second process is inside the decision boundary set in the Mahalanobis space, and determines the elevator equipment to be diagnosed as being in an abnormal state if it is outside the decision boundary.

[0021] In the present invention, the second learning model includes a GAN (Generative Adversarial Network)-based DGMR (Deep Generative Model for Regression).

[0022] In the present invention, the third learning model includes a CNN (Convolutional Neural Network).

[0023] In the present invention, the signal additionally sensed from the elevator equipment to be diagnosed is a vibration signal, and after the step of determining signs of failure of the elevator equipment to be diagnosed, the invention further comprises: a step of obtaining a vibration amount and a health weight using vibration data of the vibration signal; a step of calculating a health index (HI) using the vibration amount and the health weight; a step of accumulating and storing the health index for a set period; a step of generating a life prediction curve of the elevator equipment using the health index accumulated and stored for the set period; and a step of predicting the remaining life of the elevator equipment using the life prediction curve.

[0024] In the present invention, after the step of determining whether the elevator equipment to be diagnosed is in a normal state or an abnormal state, or the step of determining signs of failure of the elevator equipment to be diagnosed, the method further includes the step of transmitting the determination result to a worker's mobile terminal; and the step of receiving the result of the action taken on the elevator equipment from the worker at the mobile terminal and transmitting it to a server.

[0025] In addition, the AI-based elevator facility condition diagnosis system according to an embodiment of the present invention comprises a plurality of sensors that sense vibration signals or noise signals in the elevator facility; A first process of receiving the sensed signal from the sensor when the elevator equipment is in a normal state, extracting a signal in a preset frequency band, extracting feature parameters from the data of the signal, calculating the Mahalanobis distance for the extracted feature parameters, and then training the data of the Mahalanobis space defined as the data set of the Mahalanobis distances through a first learning model; and receiving an additionally sensed signal from the sensor attached to the elevator equipment to be diagnosed, extracting a signal in the frequency band, extracting feature parameters from the data of the extracted signal, calculating the Mahalanobis distance for the extracted feature parameters, and then inferring the Mahalanobis distance data in the first learning model to determine whether the elevator equipment to be diagnosed is in a normal state or an abnormal state, and if the elevator equipment to be diagnosed is determined to be in an abnormal state, generating a predicted STFT image of future data based on the current data of the additionally sensed signal from the elevator equipment to be diagnosed using a second learning model, and based on the generated predicted STFT image It includes a server that performs a second process of determining signs of failure in the elevator equipment subject to diagnosis through a third learning model.

[0026] In the present invention, the sensor is attached to a component constituting the lifting equipment and senses a vibration signal or a noise signal of the component.

[0027] In the present invention, the server oversamples data at a sampling rate higher than a preset sampling rate from the sensed signal, and filters data of the frequency band from the oversampled data to downsample it to data of the preset sampling rate.

[0028] In the present invention, the first learning model includes a One-Class Support Vector Machine (SVM), and the One-Class SVM determines the elevator equipment to be diagnosed as being in a normal state if the data of the Mahalanobis distance calculated in the second process is inside the decision boundary set in the Mahalanobis space, and determines the elevator equipment to be diagnosed as being in an abnormal state if it is outside the decision boundary.

[0029] In the present invention, the second learning model includes a GAN (Generative Adversarial Network)-based DGMR (Deep Generative Model for Regression).

[0030] In the present invention, the third learning model includes a CNN (Convolutional Neural Network).

[0031] In the present invention, the signal additionally sensed from the elevator equipment to be diagnosed is a vibration signal, and the server obtains a vibration amount and a health weight using vibration data of the vibration signal, calculates a health index (HI) using the vibration amount and the health weight, accumulates and stores the health index for a set period, generates a lifespan prediction curve of the elevator equipment to be diagnosed using the health index accumulated and stored for the set period, and predicts the remaining lifespan of the elevator equipment to be diagnosed using the lifespan prediction curve.

[0032] The present invention further includes a mobile terminal that receives a result of determining whether the elevator equipment to be diagnosed is normal or abnormal, or a result of determining signs of failure of the elevator equipment to be diagnosed, from the server, receives a result of action taken by an operator for the elevator equipment to be diagnosed, and transmits the result of action taken to the server. Effects of the invention

[0033] The AI-based elevator facility condition diagnosis system and method according to an embodiment of the present invention has one or more of the following effects.

[0034] According to the present invention, the condition of the elevator equipment can be accurately diagnosed by analyzing the mechanical condition and defect diagnosis of the elevator equipment using AI technology.

[0035] According to the present invention, signs of failure can be predicted by calculating defect characteristic values ​​from vibration and noise data measured from components of an elevator system.

[0036] According to the present invention, maintenance is convenient because the remaining lifespan of mechanical parts and major components constituting the elevator system is predicted and reported.

[0037] According to the present invention, the limitations of conventional engineering-based diagnosis are overcome, and it is possible to predict complex equipment problems that are not resolved by engineering-based diagnosis. Brief explanation of the drawing

[0038] FIG. 1 is a configuration diagram of an AI-based elevator facility condition diagnosis system according to an embodiment of the present invention. FIG. 2 is a block diagram of a server configuration according to an embodiment of the present invention. FIG. 3 is a block diagram of a data learning unit according to an embodiment of the present invention. FIG. 4 is a flowchart illustrating an AI-based elevator facility condition diagnosis method according to an embodiment of the present invention. FIG. 5 is an example of a signal sensed by a sensor according to an embodiment of the present invention and a signal extracted into a preset frequency band. FIG. 6 is an example of a Mahalanobis space for the characteristic variables of the data of a signal sensed by a sensor according to an embodiment of the present invention. FIG. 7 is a flowchart illustrating an AI-based elevator facility condition diagnosis method according to another embodiment of the present invention. FIG. 8 is an example of a crystal boundary defined in Mahalanobis space according to an embodiment of the present invention. FIG. 9 is an illustrative diagram explaining the operation of a DGMR according to an embodiment of the present invention. FIG. 10 is an illustrative diagram explaining the operation of a CNN model according to an embodiment of the present invention. FIG. 11 is a diagram illustrating the process of obtaining a soundness weight according to an embodiment of the present invention. FIG. 12 is an example of a trend line of a health index according to an embodiment of the present invention. FIG. 13 is a diagram illustrating, in its entirety, an example of an application of an AI-based elevator facility condition diagnosis system according to an embodiment of the present invention to an elevator facility. Specific details for implementing the invention

[0039] Hereinafter, some embodiments of the present invention will be described in detail with reference to exemplary drawings. It should be noted that in assigning reference numerals to the components of each drawing, the same components are given the same reference numeral whenever possible, even if they are shown in different drawings. Furthermore, in describing the embodiments of the present invention, if it is determined that a detailed description of related known components or functions would hinder understanding of the embodiments of the present invention, such detailed description is omitted.

[0040] In addition, terms such as first, second, A, B, (a), (b), etc., may be used when describing the components of the embodiments of the present invention. These terms are intended only to distinguish the components from other components, and the essence, order, or sequence of the components is not limited by the terms. Where it is stated that a component is "connected," "combined," or "connected" to another component, it should be understood that the component may be directly connected or connected to the other component, but that another component may also be "connected," "combined," or "connected" between each component.

[0041] FIG. 1 is a configuration diagram of an AI-based elevator facility condition diagnosis system according to an embodiment of the present invention.

[0042] Referring to FIG. 1, an AI-based elevator facility condition diagnosis system (10) according to an embodiment of the present invention may include a plurality of sensors (100), a server (200), and a communication network (300).

[0043] The sensor (100) is attached to the elevator equipment and can sense a signal corresponding to vibration or noise generated in the elevator equipment.

[0044] These sensors (100) may include known vibration sensors or noise sensors.

[0045] Here, lifting equipment can be a general term for equipment that moves people or cargo using an electric motor, such as elevators, escalators, and moving walkways.

[0046] In one embodiment, the sensor (100) is attached to a component constituting the elevator equipment and can sense vibration or noise generated from the component of the elevator equipment.

[0047] The components of the elevator equipment may include multiple parts and machines classified by KS, ISO standards and database information or the National Elevator Information Center.

[0048] Accordingly, preferably, the sensor (100) can be attached to a component or machine constituting the lifting equipment to sense a vibration signal or a noise signal.

[0049] The sensor (100) includes a communication module internally and can transmit the sensed vibration or noise signal to the server (200) through the communication network (300).

[0050] The communication module may include a wireless communication module such as a cellular communication module, a short-range communication module, or a GNSS (global navigation satellite system) communication module, or a wired communication module such as a LAN (local area network) communication module or a power line communication module.

[0051] The communication network (300) can handle communication between the sensor (100) and the server (200).

[0052] In one embodiment, the communication network (300) includes, but is not limited to, the Internet, LAN (Local Area Network), Wireless LAN (Wireless Local Area Network), WAN (Wide Area Network), PAN (Personal Area Network), 3G, 4G, LTE, 5G, Wi-Fi, etc.

[0053] The communication network (300) may be a closed network such as a LAN or WAN, but it is preferable that it be an open network such as the Internet.

[0054] The server (200) can process and analyze the data received from the sensor (100) to diagnose the condition of the elevator equipment and predict signs of failure and the lifespan of the elevator equipment.

[0055] The server (200) may be an intelligent server using machine learning and / or a neural network.

[0056] The server (200) can diagnose the condition of the elevator equipment by inferring the signal sensed by the sensor (100) through a preset learning model and predict signs of failure and the lifespan of the elevator equipment through a preset neural network.

[0057] FIG. 2 is a block diagram of a server configuration according to an embodiment of the present invention.

[0058] In FIG. 2, a server (200) according to an embodiment of the present invention may include a communication unit (210), a storage unit (220), an output unit (230), a weight processing unit (240), and a control unit (250).

[0059] The communication unit (210) can receive a signal from the sensor (100) through the communication network (300).

[0060] The communication unit (210) may also communicate with a mobile terminal (not shown) carried by a worker.

[0061] The storage unit (220) can store the signal received from the sensor (100).

[0062] The storage unit (220) can store the results of the condition diagnosis and fault indications of the elevator equipment from the server (200).

[0063] The storage unit (220) can store specific programs and software for executing AI-based machine learning necessary to diagnose the condition of the elevator equipment.

[0064] The weight processing unit (240) can calculate the vibration amount and soundness weight of the motor using the vibration signal data sensed from the motor of the elevator equipment among the signals sensed by the sensor (100). The weight processing unit (240) will be described in detail below.

[0065] The control unit (250) can control the overall operation of the server (200).

[0066] The control unit (250) may include an artificial intelligence processor (AI Processor) (251) for performing artificial intelligence (AI) learning using the sensing data of the sensor (100).

[0067] The AI ​​processor (251) may be configured to include a data collection unit (252) for acquiring data to be learned and a data learning unit (253) for learning the data.

[0068] The AI ​​processor (251) can learn an artificial neural network model using a program stored in the storage unit (220).

[0069] The AI ​​processor (251) can learn machine learning models and artificial neural network models for learning data related to the diagnosis of the condition of the elevator equipment using the program.

[0070] Machine learning can diagnose the condition of elevator equipment by utilizing a specific neural network. Such a neural network can be designed to simulate the structure of the human brain on a computer and may include multiple network nodes having weights that simulate neurons of the human neural network.

[0071] These multiple network modes can each exchange data to simulate the synaptic activity of neurons exchanging signals through synapses.

[0072] Neural networks can include deep learning models that have evolved from neural network models. In these deep learning models, multiple network nodes are located in different layers and can exchange data according to convolutional connection relationships.

[0073] These neural network models include various deep learning techniques such as deep neural networks (DNN), convolutional deep neural networks (CNN), recurrent Boltzmann machines (RNN), restricted Boltzmann machines (RBM), deep belief networks (DBN), and deep Q-networks, and can be applied in fields such as computer vision, speech recognition, natural language processing, and speech / signal processing.

[0074] The AI ​​processor (251) may be a general-purpose processor such as a CPU (Central Processing Unit) or AP (Application Processor), but it is preferable to implement it as an AI-dedicated processor such as an NPU (Neural Processing Unit).

[0075] The AI ​​processor (251) can learn criteria regarding which training data to use and how to classify and recognize data using the training data, acquire training data to be used for training, and train a deep learning model by applying the acquired training data to a deep learning model.

[0076] The AI ​​processor (251) may be manufactured in the form of a hardware chip and mounted on the control unit (250). For example, the AI ​​processor (251) may be manufactured in the form of a dedicated hardware chip for artificial intelligence (AI), or it may be manufactured as part of a general-purpose processor (CPU) or a graphics-dedicated processor (GPU) and mounted on the control unit (250).

[0077] The AI ​​processor (251) may be implemented as a software module. If the AI ​​processor (251) is implemented as a software module (or a program module containing instructions), the software module may be stored on a non-transitory computer-readable media. In this case, at least one software module may be provided by an operating system (OS) or by an application.

[0078] The data collection unit (252) can collect training data required for a model to classify and recognize data. For example, the data collection unit (252) can acquire data to be input into the model as training data.

[0079] The data learning unit (253) can learn to have a judgment criterion regarding how the model classifies a predetermined data using the acquired learning data.

[0080] At this time, the data learning unit (253) can train the model through supervised learning using at least some of the training data as a judgment criterion.

[0081] The data learning unit (253) may train the model through unsupervised learning, which discovers judgment criteria by learning on its own using training data without supervision.

[0082] The data learning unit (253) may train the model through reinforcement learning using feedback on whether the result of the situation judgment based on learning is correct.

[0083] The data learning unit (253) may train a neural network model using a learning algorithm including error back-propagation or gradient descent.

[0084] The data learning unit (253) can store the learned model in the storage unit (140).

[0085] FIG. 3 is a block diagram of a data learning unit according to an embodiment of the present invention.

[0086] Referring to FIG. 3, the data learning unit (253) according to an embodiment of the present invention may be configured to include a data preprocessing unit (261), a machine learning model (262), and a learning engine (263).

[0087] The data preprocessing unit (261) can generate training data by preprocessing the data of the sensed signal.

[0088] The machine learning model (262) can be composed of machine learning models capable of solving classification and decision problems, including a One-Class SVM model.

[0089] The learning engine (263) can pre-supervise the machine learning model (262) using training data (data of the sensed signal). The machine learning model (162) can learn training data containing only one type of classification data by using a One-Class SVM that utilizes a decision boundary. Additionally, supervised learning, unsupervised learning, reinforcement learning, etc., can be performed depending on the data processing method, timing, or user definition.

[0090] The learning engine (263) can perform inference of the machine learning model (162) so that the machine learning model (162) receives sensing data regarding the diagnosis of the elevator equipment and can diagnose and predict the state of the elevator equipment.

[0091] Supervised learning refers to learning that uses data containing input and corresponding output values ​​as training data to find the output value corresponding to a given input, and it means learning that takes place while the correct answer is known.

[0092] FIG. 4 is a flowchart illustrating an AI-based elevator facility condition diagnosis method according to an embodiment of the present invention, FIG. 5 is an example diagram of a signal sensed by a sensor and a signal extracted in a preset frequency band according to an embodiment of the present invention, and FIG. 6 is an example diagram of a Mahalanobis space for the characteristic variables of the data of a signal sensed by a sensor according to an embodiment of the present invention.

[0093] Referring to FIG. 4, in the AI-based elevator equipment condition diagnosis method according to an embodiment of the present invention, a vibration or noise signal is first sensed from a sensor (100) attached to an elevator equipment in a normal state (S101). This vibration or noise signal for the elevator equipment in a normal state is intended to be learned in advance through a learning model. The signal sensed by the sensor (100) can be transmitted to a server (200) through a communication network (300).

[0094] The server (200) receives a signal sensed from the sensor (100) and can extract a signal of a preset frequency band from the sensed signal (S102). This is intended to remove signals of a frequency band unnecessary for the diagnosis of the elevator equipment from the sensed signal and extract only the signals of a frequency band necessary for diagnosis. To this end, a bandpass filter is used as a Finite Impulse Response (FIR) filter to extract only the signals of the preset frequency band.

[0095] In the signal sensed by the actual sensor (100), an aliasing phenomenon may occur in which high-frequency components are mixed with low-frequency components, and this phenomenon may cause signal distortion. Therefore, in this embodiment, to reduce the distortion of the sensed signal, the data is oversampled at a sampling rate much higher than the preset sampling rate, and then, using bandpass filtering as an FIR filter, only the signal of the preset frequency band is retained, and then downsampled to extract the signal of the desired frequency band. FIG. 5 illustrates an example in which the signal sensed by the sensor (100) and the signal of the preset frequency band are extracted.

[0096] Next, the server (200) can extract feature parameters from the data of the signal in a specific frequency band extracted as described above (S103). Here, the feature parameters may refer to each variable or attribute observed in the data of the signal in the extracted frequency domain. These feature parameters are numerical representations for extracting meaningful information from raw data and are key elements that help machine learning models understand and predict the data. In the field of vibration and noise, features can be extracted through various statistics and energy-based indicators in the time, frequency, and time-frequency domains.

[0097] For example, to classify the above data, the size, frequency, location, and presence or absence of impact of the data may be used as features. In this case, feature variables may be used to identify the most useful or important characteristic among these characteristics.

[0098] The server (200) can calculate the Mahalanobis distance for the extracted feature variables (S104). The Mahalanobis distance is calculated to determine how far each data point is from the steady-state data range in multidimensional space. Compared to the general Euclidean distance, it can ensure statistical reliability by considering the covariance. Since deviations exist even among the feature variables of each signal, data on how the steady-state data is distributed is obtained by considering the deviations as well.

[0099] Figure 6 illustrates, as an example, the distribution of Mahalanobis distances calculated for feature variables of sensed data.

[0100] Subsequently, the server (200) can perform learning on the data of the Mahalanobis Space (MS), which is the data set of the calculated Mahalanobis distances, through a preset first learning model (S105).

[0101] In this embodiment, the first learning model may include a One-Class Support Vector Machine (SVM). The One-Class SVM can be a learning model for determining whether the Mahalanobis distance data is normal state data or abnormal state data, and defines a Mahalanobis space for this determination.

[0102] FIG. 7 is a flowchart illustrating an AI-based elevator facility condition diagnosis method according to another embodiment of the present invention, FIG. 8 is an example of a decision boundary defined in Mahalanobis space according to an embodiment of the present invention, FIG. 9 is an example illustrating the operation of DGMR according to an embodiment of the present invention, and FIG. 10 is an example illustrating the operation of a CNN model according to an embodiment of the present invention.

[0103] Referring to FIG. 7, in an AI-based elevator equipment condition diagnosis method according to another embodiment of the present invention, the first learning model described in FIG. 4 is used to predict signs of failure and lifespan of the elevator equipment by using additionally sensed signals.

[0104] With the first learning model established as described above, a vibration or noise signal of the elevator equipment is additionally sensed by a sensor (100) attached to the elevator equipment (S201). The additionally sensed signal (hereinafter referred to as the additional sensing signal) is transmitted to a server (200).

[0105] The first learning model above is a One-Class SVM, and as described above, a Mahalanobis space is defined for the signals sensed in the steady-state elevator equipment, and a decision boundary is set.

[0106] The server (200) receives an additional sensing signal from the sensor (100) and can extract a signal of a preset frequency band from the additional sensing signal (S202). The extraction of such a signal can be performed in the same way as step S102 described above.

[0107] Next, the server (200) extracts feature variables from the data of the signal of a specific frequency band extracted as described above (S203). Then, it calculates the Mahalanobis distance for the extracted feature variables (S204).

[0108] The server (200) diagnoses the condition of the elevator equipment by inferring the data of the Mahalanobis distance from the first learning model, such as One-Class SVM, which is already established in FIG. 4 (S205).

[0109] To this end, it is checked whether the data of the above Mahalanobis distance is inside or outside the preset decision boundary.

[0110] If the data of the above Mahalanobis distance is inside the decision boundary set in the Mahalanobis space, the elevator equipment is determined to be in a normal state, and conversely, if it is outside the decision boundary, the elevator equipment is determined to be in an abnormal state.

[0111] In other words, if the Mahalanobis distance of the newly input additional sensing signal falls within the decision boundary of the pre-trained One-Class SVM model, the corresponding elevator equipment is determined to be in a normal state, and if it falls outside the decision boundary, it is determined to be in an abnormal state.

[0112] Figure 8 is an example of a decision boundary set in the Mahalanobis space. As shown in Figure 8, data inside the decision boundary is determined to be normal state data, and data outside the decision boundary is determined to be abnormal state data.

[0113] In other words, in this embodiment, data inside the determination boundary is determined to indicate that the elevator equipment is in a normal state, and the opposite is determined to indicate that the elevator equipment is in an abnormal state.

[0114] Subsequently, if it is determined that the elevator equipment is in an abnormal state (S206), a prediction STFT (Short Time Fourier Transform) image of the additional sensing signal data can be generated using the second learning model (S207).

[0115] In this embodiment, such predictive STFT images can be used to predict future data based on current data. Specifically, in order to calculate defect features from vibration or noise signal data measured in elevator equipment, analyze trends, and predict signs of failure, it is important to analyze the trends of defect feature values.

[0116] Accordingly, in this embodiment, a second learning model is used to generate a predicted STFT image of future data in order to identify future changes based on current data sensed by the elevator equipment. At this time, the second learning model may include a Deep Generative Model for Regression (DGMR) model based on a Generative Adversarial Network (GAN).

[0117] A GAN is a model composed of two neural networks, namely a generator (G) and a discriminator (D). The generator (G) creates new data, while the discriminator (D) determines whether the generated data is real or fake. This enables the generation of increasingly realistic data samples.

[0118] The DGMR model is a variant of the GAN designed to solve regression problems. Because it can model continuous changes in data over time, it is suitable for predicting time-series data such as vibration or noise. Consequently, when predicting signs of failure, it can be used to forecast future data and compare it with current data.

[0119] FIG. 9 illustrates the operation of generating future predicted STFT images in a DGMR model. Referring to FIG. 9, in DGMR, Observation (301) refers to actual data from various points in time obtained through observation, and Context (302) is past information or conditional input, which is a condition required for the model to generate the future. For example, if an STFT image of the first 5 seconds of sensor time series is input, the Generator (303) thereafter predicts the next time or product based on the Context (302). To verify the naturalness, consistency, and consistency of the generated image, the future predicted STFT image is generated after undergoing verification evaluations within DGMR, such as Temporal consistency, Spatial consistency, and Sample consistency.

[0120] Subsequently, the third learning model can determine signs of failure in the elevator equipment using the predicted STFT image (S208). Preferably, signs of failure in a component or mechanical device constituting the elevator equipment are determined.

[0121] The above third learning model may include a Convolutional Neural Network (CNN) model. Preferably, it may include a CNN-based fault indication model.

[0122] CNNs are deep learning models specialized in analyzing high-dimensional data such as images, and are primarily used for image classification, object detection, and feature extraction of time-series data. By inputting the STFT images of the previously generated prediction data into these models, signs of future failure are identified.

[0123] Figure 10 illustrates the overall operation of fault detection in a CNN model. Referring to Figure 10, the previously obtained future prediction STFT image undergoes an image feature extraction process through a CNN. Patterns are extracted in the Convolutional Layer, followed by resolution reduction and feature summarization through the Max Pooling process. The Convolutional Layer and Max Pooling processes are repeated as many times as defined by the model design. Subsequently, as a feature classification step, the FC Layers output the results for fault detection.

[0124] Here, although not illustrated in the drawings, in an embodiment of the present invention, the results of the condition diagnosis and fault indications of the elevator equipment may be transmitted to a worker's mobile terminal or displayed on a display device.

[0125] The operator can take measures such as maintaining, repairing, or replacing parts of the elevator equipment by viewing the results received on their mobile terminal. Even after taking such measures, the sensor (100) attached to the elevator equipment can measure vibration or noise signals in real time. Then, using the signals measured in real time in this way, the condition of the elevator equipment and signs of failure can be determined as shown in FIGS. 4 and FIGS. 7.

[0126] In the present invention, diagnosis of each component and mechanical device constituting the elevator system is not merely a determination of whether it is in a normal or abnormal state, but enables comprehensive judgments such as signs of failure and lifespan prediction. To achieve this, data from the elevator system maintenance company is collected, and the overall status of the system regarding the corrective actions taken is gathered via sensors, and modeling can be performed based on this. Therefore, the present invention can provide a prediction service capable of making comprehensive judgments about the elevator system through a trained model. In particular, by utilizing deep learning, feature factors can be autonomously extracted from training data, and based on this, classification and prediction problems can be solved.

[0127] Engineering-based prediction algorithms are based on Bayes' Rule, which explains how to calculate the probability of event B occurring when event A has already occurred, and make predictions based on judgments derived from the opinions of maintenance managers diagnosing equipment conditions in the field.

[0128] While engineering-based prediction algorithms can intuitively diagnose individual components, they struggle to predict unconsidered causes or complex equipment issues.

[0129] CNN, a deep learning technique, is used to address the difficulty of such predictions. After preprocessing to visualize time-series signals such as vibration or noise signals, failure signs can be predicted without relying on the empirical judgment of maintenance managers through the analysis of diverse and complex data.

[0130] Meanwhile, the present invention can predict the remaining lifespan of an elevator system. In particular, it predicts the remaining lifespan of major components constituting the elevator system, such as electric motors, reduction gears, and bearings, using vibration data. Below, an electric motor is described as an example.

[0131] As a typical example of configuring the lifting equipment, a sensor (100) is attached to the motor to sense the vibration signal of the motor. The sensed signal is transmitted to a server (200).

[0132] The server (200) calculates the vibration amount and soundness weight by using the 3-axis direction vibration data of the vibrator in the weight processing unit (240) and going through a preprocessing process and a Fast Fourier Transform (FFT) process.

[0133] In the preprocessing of vibration data, the magnitude of a 3D vector is extracted from the 3-axis vibration data, and the DC component corresponding to gravitational acceleration is removed from the extracted magnitude of the 3D vector.

[0134] In the Fast Fourier Transform (FFT) process, the signal output from the preprocessing stage is subjected to an FFT transformation to extract frequencies for each dimension.

[0135] The weight processing unit (240) calculates the vibration amount and soundness weight using the vibration data that has undergone the above preprocessing process and FFT transformation.

[0136] The vibration amount is calculated using the frequency values ​​of each dimension in the frequency domain of the vibration data. The evaluation of an electric motor's condition using vibration amount is defined in ISO 10816 and is widely used as a condition indicator demonstrating decades of reliability in general electric motor condition diagnosis. In this embodiment, the vibration amount (Overall) can be calculated as shown in Equation 1 below.

[0137]

[0138] Here, Overall(OA RSS ) is the amount of vibration of the motor, and A1, A2, ..., An are the frequency values ​​of each dimension in the Fast Fourier Transform.

[0139] Soundness weighting ( ) is a weighting factor that adjusts how reliable the motor's failure or lifespan is with respect to the motor's vibration data.

[0140] In this embodiment, a CNN-based weight calculation model is used to calculate the health weight. Specifically, vibration data is visualized, and the visualized vibration data is transformed using the recurrence plot technique. The transformed recurrence plot is then applied to the CNN-based weight calculation model to obtain the health weight. The CNN-based model is a model that demonstrates strong characteristics in recognizing image features.

[0141] The recurrence plot technique is an effective method for identifying and visualizing periodicity or trajectories in signals recording motion, and it is useful for verifying signal periodicity in general scientific analysis. In this embodiment, the recurrence plot technique generates an image of vibration data that includes features related to the motor's health (or failure) contained within the motor's vibration data signal. In other words, the vibration data is converted into an image that includes features related to the motor's health.

[0142] These vibration data images are converted so that they can be applied to a CNN-based model. Generally, since vibration data signals exhibit periodicity when detected by an electric motor, the imaging of the vibration data allows the CNN-based model to extract health (failure)-related features from the signals, perform evaluation and training, and obtain data containing the motor's health (failure) characteristics included in the images.

[0143] FIG. 9 is a diagram illustrating the process of obtaining a soundness weight according to an embodiment of the present invention.

[0144] Referring to FIG. 9, vibration data (401) of an electric motor is acquired through a sensor (100), and an image (402) of the vibration data is generated by transforming the vibration data (401) using a recurrence plot technique. Subsequently, the image (402) of the vibration data is applied as an input value to a CNN-based health weight calculation model to calculate the health weight.

[0145] The health weighting factor has a value between 0 and n. A value closer to 0 indicates that the motor is in good condition and has a relatively long lifespan, while a value further from 0 indicates that the motor is in poor condition and has a relatively short lifespan. Here, n is defined in advance by the user based on the type and condition of the motor for which the remaining lifespan is to be predicted.

[0146] The control unit (250) can calculate the Health Index of the motor using the vibration amount and health weight of the motor.

[0147] To predict the remaining lifespan of an electric motor, a factor capable of comprehensively assessing its condition is required. This is called the Health Index (HI), and the soundness of the motor's components can be determined based on the magnitude of the Health Index.

[0148] The control unit (250) calculates a value obtained by multiplying a vibration amount, which is frequently used for evaluating the motor, by a health weighting factor having a value between 0 and n, as shown in the following mathematical formula 2, and uses it as a health index.

[0149]

[0150] The control unit (250) can calculate the health index (HI) and store the calculated health index in the internal storage unit (220).

[0151] In predicting the remaining lifespan of an electric motor, if there is a significant difference in the motor's state changes, it is possible to estimate the lifespan prediction curve from the trend of that difference.

[0152] Typically, it takes at least 6 months to 1 year to obtain a significant change in the state of the motor. Accordingly, the control unit (250) acquires the health index for the set period daily and accumulates and stores it in the storage unit (220).

[0153] The control unit (250) generates a life prediction curve for the motor using a health index accumulated in the storage unit (220) for a certain period of time and calculates the remaining life of the motor using the life prediction curve. Specifically, the failure of the motor occurs as factors leading to the failure accumulate over time, and these factors can be reflected in vibration.

[0154] To this end, the control unit (250) calculates the health index trend line after the above-mentioned time by applying an index reduction model that uses a health index reflecting these elements as an input value.

[0155] The exponential degradation model applied in this embodiment is a mathematically based model that predicts trends over subsequent time using input values ​​calculated over a certain period, and it is a model that assumes that the numerical value of the state change over time of an arbitrary observed object appears in the form of an exponential curve. In other words, it models the process in which the previous states of an arbitrary observed object continuously accumulate to influence the change in the current state, and expresses this process using a mathematical formula. Typically, changes in the state related to the lifespan of multiple mechanical and electronic devices show a trend that largely follows this assumption. The formula for the exponential degradation model can generally be represented by Equation 3.

[0156]

[0157] Here, A(ti) is the formula (function) of the exponential resistance model with respect to time t, Φ is the intercept of the exponential resistance model, θ and β are the random variables of the log-normal and Gaussian normal distributions based on the mean and variance of the accumulated health index, and ε(ti) and σ2 are the random variables of the normal distribution of noise and the variance of the noise that the exponential resistance model can have.

[0158] In the above mathematical formula 3, the cumulative value of the health index (the health index and the time at which the health index was calculated) is used to calculate the variables of each phase. The constant and variable values ​​of each term are calculated using the cumulative value of the health index, and the calculation is repeated until the value of ti (i=0,…,n) is approximated by the user-specified expected health index. The sequence of the expected health index and the sequence of ti, which are the calculated values ​​obtained by this repetition, are used as the lifespan prediction curve of the health index.

[0159] FIG. 10 illustrates an example of a trend line of a health index according to an embodiment of the present invention. First, the health index indicated by A is a health index calculated over a certain period of time (e.g., from July 2021 to January 2022), and the health index indicated by B is a predicted value of the health index for a subsequent period by applying the health indices (A) over the said certain period of time to an index degradation model. In this way, the health index accumulated over a certain period is applied as an input value to the index degradation model to predict the health index for a period after the said certain period, and the health indices predicted for the said certain period and the subsequent period are combined to generate a lifespan prediction curve.

[0160] Afterward, the control unit (250) calculates the remaining lifespan as the time until the point in time when the health index threshold (C) determined to be a failure is reached in the health index prediction graph (B). In the example of FIG. 10, the motor is predicted to be usable until April 2047.

[0161] These lifespan prediction results can be displayed on a display device.

[0162] FIG. 11 is a diagram illustrating, in its entirety, an example in which an AI-based elevator facility condition diagnosis system and method according to an embodiment of the present invention are applied to an elevator facility.

[0163] Sensors are attached to the elevator equipment, preferably to the various parts and mechanical devices constituting the elevator equipment, so that the sensors detect vibration or noise signals from each part and mechanical device. The signals detected in this way are input into an elevator equipment condition diagnosis system.

[0164] The elevator equipment condition diagnosis system diagnoses whether each component and mechanical device is in a normal or abnormal state, and if diagnosed as abnormal, it additionally predicts signs of failure and remaining lifespan of the relevant component and mechanical device.

[0165] These diagnostic and prediction results can be transmitted to a mobile terminal carried by the worker. Based on the information transmitted to the mobile terminal, the worker takes measures such as maintenance, repair, or replacement of the lifting equipment, individual parts, and machinery.

[0166] When the elevator system is operated following these measures, vibration or noise signals are detected again in real time by each sensor. In other words, signals are sensed from the elevator system that reflects the results of the measures. These signals are also input into the elevator system status diagnosis system, where diagnosis and prediction are repeated.

[0167] As described above, the AI-based elevator equipment condition diagnosis system according to an embodiment of the present invention introduces not only engineering-based diagnostic technology but also artificial intelligence (AI) diagnostic technology to diagnose the condition and defects of mechanical parts and major components constituting the elevator equipment. This minimizes numerous variables required for diagnosis and enables the prediction and resolution of complex problems in the equipment. Furthermore, the present invention reduces the discrepancy between the empirical diagnostic results of maintenance managers and the AI-based data diagnostic results by reflecting the inspection results and history—such as maintenance results, parts replacement, and fault handling—obtained through on-site inspections based on the diagnostic results of the elevator equipment back into the analysis unit.

[0168] In the case of engineering-based diagnosis generally used in the field of elevator equipment diagnosis, diagnosis is performed based on mechanical analysis and diagnostic theories derived from the unique design specifications of each piece of equipment. However, since the installation environment varies for each piece of equipment and it is necessary to establish fixed rules considering numerous variables such as lifespan and mechanical characteristics, it is difficult to ensure system flexibility and there are disadvantages such as the difficulty in selecting appropriate thresholds during equipment diagnosis. In contrast, the present invention aims to overcome these conventional limitations by introducing diagnostic technology utilizing artificial intelligence (AI), thereby enabling the prediction of complex elevator equipment problems that cannot be resolved by engineering-based diagnosis.

[0169] In addition, while existing systems employing artificial intelligence diagnostic technology rely on a method of training by processing measured data and manually labeling it through human intervention, the AI-based elevator condition diagnostic system of the present invention adopts a method of automatically labeling field inspection results by considering the experiential judgment of on-site maintenance managers; based on this, retraining is performed to enhance the system's suitability for the field.

[0170] Accordingly, the AI-based elevator equipment condition diagnosis system according to the present invention diagnoses the overall condition and defects of elevators, escalators, moving walkways, etc., based on engineering and artificial intelligence, predicts the future condition and remaining lifespan of the target equipment, and incorporates field inspection results into the learning process; thus, it assists in effective maintenance and efficient equipment management and has the advantage of enhancing passenger safety.

[0171] In the foregoing, although all components constituting an embodiment of the present invention have been described as being combined or operating in combination, the present invention is not necessarily limited to such embodiments. That is, within the scope of the purpose of the present invention, all components may be selectively combined in one or more ways to operate. Furthermore, terms such as "include," "constitute," or "have" described above, unless specifically stated otherwise, mean that the relevant component may be inherent; thus, they should be interpreted as allowing for the inclusion of additional components rather than excluding other components. All terms, including technical or scientific terms, have the same meaning as generally understood by those skilled in the art to which the present invention pertains, unless otherwise defined. Terms commonly used, such as those defined in advance, should be interpreted in accordance with their meaning in the context of the relevant technology and, unless explicitly defined in the present invention, should not be interpreted in an ideal or overly formal sense.

[0172] The foregoing description is merely an illustrative explanation of the technical concept of the present invention, and those skilled in the art to which the present invention pertains will be able to make various modifications and variations within the scope of the essential characteristics of the present invention. Accordingly, the embodiments disclosed in the present invention are intended to explain, not limit, the technical concept of the present invention, and the scope of the technical concept of the present invention is not limited by such embodiments. The scope of protection of the present invention shall be interpreted by the claims below, and all technical concepts within an equivalent scope shall be interpreted as being included within the scope of rights of the present invention. Explanation of the symbols

[0173] 100: Sensor 200: Server 210: Communications Unit 220: Storage Unit 230: Output unit 240: Weight processing unit 250: Control unit 251: AI processor 252: Data Collection Department 253: Data Learning Department 261: Preprocessing Section 262: Machine Learning Model 263: Learning Engine 300: Communication Network

Claims

Claim 1 (a) sensing a vibration or noise signal in a normal state elevator; (b) extracting a signal of a preset frequency band from the sensed signal; (c) extracting feature parameters from the data of the extracted signal; (d) calculating a Mahalanobis distance for the feature parameters; (e) training a One-Class Support Vector Machine (SVM) model based on data of the Mahalanobis space defined by the data set of the calculated Mahalanobis distances to set a decision boundary for the Mahalanobis space; (f) additionally sensing a vibration or noise signal in the elevator of the diagnosis target; (g) extracting a signal of the frequency band from the additionally sensed signal; (h) extracting feature parameters from the data of the extracted signal; (i) calculating the Mahalanobis distance for the feature parameters; (j) training the One-Class SVM model based on the calculated Mahalanobis distance A step of determining whether the elevator equipment to be diagnosed is in a normal state or an abnormal state based on whether the calculated Mahalanobis distance inferred from the SVM model is inside or outside the decision boundary set in step (e); (k) if the elevator equipment to be diagnosed is determined to be in an abnormal state, a step of generating a predicted STFT image of future data based on current data of additionally sensed signals from the elevator equipment to be diagnosed using a pre-trained DGMR (Deep Generative Model for Regression) model;and (l) a step of determining signs of failure in the elevator equipment to be diagnosed through a pre-trained CNN (Convolutional Neural Network) model based on the generated predicted STFT image, wherein step (b) actively extracts a signal of the desired frequency band by oversampling the data at a sampling rate higher than a preset sampling rate, leaving only the signal of the preset frequency band through bandpass filtering, and then downsampling; and step (j) determines the elevator equipment to be diagnosed as being in a normal state if the Mahalanobis distance calculated in step (i) is inside the decision boundary set in the Mahalanobis space defined in the pre-trained One-Class SVM model in step (e), and determines the elevator equipment to be diagnosed as being in an abnormal state if it is outside the decision boundary; and the CNN model extracts the pattern of the predicted STFT image of the future data obtained in step (k) in the Convolutional Layer, reduces the resolution of the extracted pattern and extracts features through a Max Pooling process, and then again combines the Convolutional Layer and Max Pooling An AI-based elevator equipment condition diagnosis method that proceeds through the process for a predetermined definition in the model and outputs a prediction explanation result for fault detection in the FC Layers. Claim 2 In claim 1, the step of sensing a vibration or noise signal of the elevator equipment is an AI-based elevator equipment condition diagnosis method in which a sensor attached to a component constituting the elevator equipment senses a vibration signal or noise signal of the component. Claim 3 The AI-based elevator equipment condition diagnosis method according to claim 1, wherein the step of extracting a signal of the frequency band excludes signals unnecessary for the diagnosis of the elevator equipment and filters only the signals necessary for the diagnosis of the elevator equipment. Claim 4 delete Claim 5 delete Claim 6 delete Claim 7 delete Claim 8 delete Claim 9 The AI-based elevator condition diagnosis method according to claim 1, wherein the signal additionally sensed in the elevator equipment to be diagnosed is a vibration signal, and after the step of determining signs of failure of the elevator equipment to be diagnosed, the method further comprises: a step of obtaining a vibration amount and a health weight using vibration data of the vibration signal; a step of calculating a health index (HI) using the vibration amount and the health weight; a step of accumulating and storing the health index for a set period; a step of generating a life prediction curve of the elevator equipment using the health index accumulated and stored for the set period; and a step of predicting the remaining life of the elevator equipment using the life prediction curve. Claim 10 The AI-based elevator equipment condition diagnosis method according to claim 1, further comprising: a step of determining whether the elevator equipment to be diagnosed is in a normal state or an abnormal state, or a step of determining signs of failure of the elevator equipment to be diagnosed, a step of transmitting the determination result to a worker's mobile terminal; and a step of receiving the result of the action taken on the elevator equipment from the worker at the mobile terminal and transmitting it to a server. Claim 11 Multiple sensors for sensing vibration signals or noise signals in elevator equipment; A first process of receiving a sensed signal from the sensor when the elevator equipment is in a normal state, extracting a signal in a preset frequency band, extracting feature parameters from the data of the extracted signal, calculating a Mahalanobis distance for the extracted feature parameters, and then training a One-Class SVM (Support Vector Machine) model based on the data of the Mahalanobis space defined by the data set of the calculated Mahalanobis distances to set a decision boundary for the Mahalanobis space; and receiving an additional sensed signal from the sensor attached to the elevator equipment to be diagnosed, extracting a signal in the frequency band, extracting feature parameters from the data of the extracted signal, calculating a Mahalanobis distance for the extracted feature parameters, and then inferring the data of the Mahalanobis distance from the trained One-Class SVM model to determine whether the calculated Mahalanobis distance is inside or outside the decision boundary set in step (e), based on the diagnosis target A server that determines whether the elevator equipment is in a normal state or an abnormal state, and if the elevator equipment to be diagnosed is determined to be in an abnormal state, generates a predicted STFT image of future data based on current data of signals additionally sensed from the elevator equipment to be diagnosed using a pre-trained DGMR (Deep Generative Model for Regression) model, and performs a second process of determining signs of failure of the elevator equipment to be diagnosed through a pre-trained CNN (Convolutional Neural Network) model based on the generated predicted STFT image;The method includes the step of receiving the sensed signal from the sensor and extracting a signal of a preset frequency band, wherein the data is oversampled at a sampling rate higher than the preset sampling rate, and then, after leaving only the signal of the preset frequency band through bandpass filtering, the desired signal of the said frequency band is actively extracted by downsampling; the step of determining whether the elevator equipment to be diagnosed is in a normal state or an abnormal state is wherein the elevator equipment to be diagnosed is determined to be in a normal state if the calculated Mahalanobis distance is inside the decision boundary set in the Mahalanobis space defined in the pre-trained One-Class SVM model, and the elevator equipment to be diagnosed is determined to be in an abnormal state if it is outside the decision boundary; and the CNN model extracts the pattern of the predicted STFT image of the future data in the Convolutional Layer, reduces the resolution of the extracted pattern and extracts features through a Max Pooling process, performs the Convolutional Layer and Max Pooling process again for a number of times defined in the model, and outputs the prediction explanation result for fault detection in the FC Layers, thereby providing an AI-based elevator equipment state. Diagnostic system.; Claim 12 In claim 11, the sensor is attached to a component constituting the elevator equipment and senses a vibration signal or noise signal of said component in an AI-based elevator equipment condition diagnosis system. Claim 13 delete Claim 14 delete Claim 15 delete Claim 16 delete Claim 17 An AI-based elevator condition diagnosis system according to claim 11, wherein the signal additionally sensed from the elevator equipment to be diagnosed is a vibration signal, the server obtains a vibration amount and a health weight using vibration data of the vibration signal, calculates a health index (HI) using the vibration amount and the health weight, accumulates and stores the health index for a set period, generates a life prediction curve of the elevator equipment to be diagnosed using the health index accumulated and stored for the set period, and predicts the remaining life of the elevator equipment to be diagnosed using the life prediction curve. Claim 18 An AI-based elevator condition diagnosis system according to claim 11, further comprising a mobile terminal that receives a result of determining whether the elevator equipment to be diagnosed is normal or abnormal, or a result of determining signs of failure of the elevator equipment to be diagnosed, from the server, receives a result of action taken by an operator for the elevator equipment to be diagnosed, and transmits the result of action taken to the server.

Images