Smart Factory Fault Prediction System and Fault Prediction Method

Abstract

The fault prediction system and method for a smart factory according to an embodiment of the present invention have the advantage of accurately converting abnormal behavior patterns of equipment directly linked to product defects into data by precisely synchronizing the time of quality judgment of a product through vision inspection with vibration and sound data, which are physical state signals of the equipment, based on timestamps.

Description

Technology Field

[0001] The present invention relates to a fault prediction system and a fault prediction method for a smart factory. Background Technology

[0003] Generally, production facilities in the manufacturing sector are exposed to high temperatures, vibrations, and high voltages, or perform repetitive manufacturing processes. Therefore, a maintenance process is essential to replace worn or old parts and check for failures or defects by performing preventive maintenance at regular intervals.

[0004] In particular, the introduction of automated equipment, which is unmanned automated equipment, is increasing due to factors such as the recent decline in workforce, aging population, and reduced productivity in the manufacturing sector. By performing proper maintenance before breakdowns or defects occur in such automated equipment, it is possible to reduce cost losses due to breakdowns, extend the lifespan of the automated equipment, and minimize losses caused by production line interruptions.

[0005] Meanwhile, conventionally, maintenance is performed at regular intervals regardless of whether actual defects or breakdowns have occurred in automated equipment. Consequently, automated equipment without actual defects is included in the maintenance targets, leading to the problem of unnecessary consumption of time and manpower.

[0006] In addition, there is a problem in that parts are replaced unnecessarily during the maintenance of automated equipment, or breakdowns occur in the automated equipment before the maintenance time, resulting in high maintenance costs.

[0007] Recently, devices that collect information such as equipment vibrations and use this to predict failures in advance are being developed.

[0008] However, conventional fault prediction devices are systems that simply determine a fault when vibrations exceed upper and lower limits, and since the results from this prediction method are not precise, it is difficult to accurately determine the presence or absence of a fault. Prior art literature

[0010] Republic of Korea Registered Patent No. 10-0347008 Republic of Korea Registered Patent No. 10-1280802 Republic of Korea Registered Patent No. 10-1526100 The problem to be solved

[0011] The present invention aims to provide a fault prediction system and a fault prediction method for a smart factory. means of solving the problem

[0013] The present invention comprises a data acquisition step (S10) in which multidimensional process data is collected in real time through a vision detection unit (110), a vibration detection unit (130), and a sound detection unit (150) of a detection device (100) simultaneously with the operation of an automated facility (10); a vision inspection step (S20) in which the vision inspection unit (120) analyzes the vision data and determines whether the product (20) is good or bad; a data integration step (S30) in which, when a defective product is determined in the vision inspection step (S20), a control unit (170) generates a timestamp at that time and matches the vibration data and sound data synchronized with it to generate defect indication data; a data preprocessing step (S40) in which the data preprocessing module (180) removes noise from the defect indication data, unifies the data specifications, and transmits them to a fault prediction server (30); and an AI learning engine (220) of the fault prediction server (30) detects equipment failure based on the accumulated defect indication data. A control method for a fault prediction system of a smart factory is provided, comprising an AI learning step (S50) for optimizing a fault prediction algorithm by learning a pattern, and a similarity analysis step (S60) in which the comparison judgment module (250) calculates a similarity by comparing real-time incoming vibration and sound data with the learned fault prediction algorithm.

[0014] If the similarity calculated in the above similarity analysis step (S60) is less than 90%, it may include a failure prediction notification step (S70) that outputs a preventive maintenance notification by determining that it is a sign of failure of the automated equipment (10) before the actual failure occurs.

[0015] The fault prediction system of a smart factory according to the present invention may include a detection device (100) that is physically connected to and installed with an automation facility (10) and generates and extracts data from the automation facility (10), and a fault prediction server (30) that is connected to the detection device (100) via a wired or wireless communication network and receives and processes fault indication data. Effects of the invention

[0017] The effects of the fault prediction system and fault prediction method of a smart factory according to an embodiment of the present invention are as follows.

[0018] First, the present invention has the advantage of accurately digitizing abnormal behavior patterns of equipment directly linked to product defects by precisely synchronizing the timing of product quality judgment through vision inspection with vibration and sound data, which are physical state signals of the equipment, based on timestamps.

[0019] Second, the present invention utilizes a piezoelectric vibration detection sensor and a high-sensitivity microphone array to capture even minute signals in the ultrasonic band as well as the audible band, thereby having the advantage of precisely predicting early defects in equipment that are difficult to detect using conventional simple threshold methods.

[0020] Third, the present invention has the advantage of increasing the efficiency of artificial intelligence learning and improving the judgment reliability of the fault prediction algorithm by removing on-site electrical noise and environmental noise through a data preprocessing module and unifying data specifications.

[0021] Fourth, the present invention has the advantage of enabling intuitive and effective failure prediction even amidst complex external process variables such as temperature, humidity, and raw materials, as it learns by directly matching the equipment output value at the time of production, rather than the root cause of the defect.

[0022] Fifth, the present invention has the advantage of effectively preventing equipment damage and blocking the mass production of defects by providing a step-by-step control process that notifies of failure signs in advance through real-time similarity analysis and automatically stops the equipment when the defect rate is exceeded.

[0023] Sixth, the present invention has the advantage of maintaining the accuracy of fault prediction even in the harsh process environment of a smart factory by increasing the purity of vibration signals transmitted through the equipment structure by installing the vibration detection module spaced apart from the column via a module mount. Brief explanation of the drawing

[0025] FIG. 1 is a perspective view illustrating a fault prediction system of a smart factory according to a first embodiment of the present invention. FIG. 2 is a block diagram of a fault prediction system for a smart factory according to a first embodiment of the present invention. Figure 3 is a schematic cross-sectional view of the vibration sensing unit shown in Figure 1. FIG. 4 is a flowchart illustrating a control method of a fault prediction system of a smart factory according to a first embodiment of the present invention. Specific details for implementing the invention

[0026] The present invention is susceptible to various modifications and may have various embodiments; specific embodiments are illustrated in the drawings and described in detail in the detailed description. However, this is not intended to limit the invention to specific embodiments, and it should be understood that the invention includes all modifications, equivalents, and substitutions that fall within the spirit and scope of the invention. Similar reference numerals have been used for similar components in the description of each drawing.

[0027] Terms such as first, second, A, B, etc., may be used to describe various components, but said components shall not be limited by said terms. These terms are used solely for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, the first component may be named the second component, and similarly, the second component may be named the first component. The term "and / or" includes a combination of a plurality of related described items or any of a plurality of related described items.

[0028] When it is stated that one component is "connected" or "connected" to another component, it should be understood that while it may be directly connected or connected to that other component, there may also be other components in between. On the other hand, when it is stated that one component is "directly connected" or "directly connected" to another component, it should be understood that there are no other components in between.

[0029] The terms used in this application are used merely to describe specific embodiments and are not intended to limit the invention. The singular expression includes the plural expression unless the context clearly indicates otherwise. In this application, terms such as "comprising" or "having" are intended to specify the presence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.

[0030] Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as generally understood by those skilled in the art to which the present invention pertains. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and should not be interpreted in an ideal or overly formal sense unless explicitly defined in this application.

[0031] In this document, "configured to" may be used interchangeably with, depending on the context, for example, in hardware or software, "suitable for," "capable of," "modified to," "made to," "capable of," or "designed to." In some situations, the expression "device configured to" may mean that the device is "capable of" doing something together with other devices or components.

[0032] Hereinafter, preferred embodiments of the present invention will be described in more detail with reference to the attached drawings. In order to facilitate an overall understanding of the present invention, the same reference numerals are used for identical components in the drawings, and redundant descriptions of identical components are omitted.

[0033] FIG. 1 is a perspective view illustrating a fault prediction system of a smart factory according to a first embodiment of the present invention, FIG. 2 is a block diagram of a fault prediction system of a smart factory according to a first embodiment of the present invention, FIG. 3 is a schematic cross-sectional view of a vibration detection unit illustrated in FIG. 1, and FIG. 4 is a flowchart illustrating a control method of a fault prediction system of a smart factory according to a first embodiment of the present invention.

[0034] The fault prediction system of a smart factory according to the present embodiment includes a detection device (100) that is physically connected to and installed with an automation facility (10) and generates and extracts data from the automation facility (10), and a fault prediction server (30) that is connected to the detection device (100) via a wired / wireless communication network and receives and processes fault indication data.

[0035] The detection device (100) of the fault prediction system of a smart factory according to the present embodiment includes a vision detection unit (110) that captures an image of a product (20) produced in an automated facility (10), a vision inspection unit (120) that extracts vision data from the image captured by the vision detection unit (110) and determines whether the product is a defective product or a normal product through the extracted vision data, a vibration detection unit (130) that detects vibration of the automated facility (10), a vibration inspection unit (140) that converts a signal detected by the vibration detection unit (130) into vibration data and determines abnormal vibration from the vibration data, and a control unit (170) that is electrically connected to the vision inspection unit (120) and the vibration inspection unit (140) and processes the vision data and vibration data.

[0036] The above vision inspection unit (120) includes a vision data processing unit, a vision image acquisition unit, and a vision defect determination unit. The vision inspection unit (120) may include a configuration (e.g., a robot arm) for separating a product determined to be defective through vision inspection from an automated line.

[0037] The above vibration inspection unit (140) includes a vibration data processing unit, a vibration data acquisition unit, and an abnormal vibration determination unit.

[0038] The above sound inspection unit (160) includes a sound data processing unit, a sound data acquisition unit, and an abnormal sound determination unit.

[0039] The detection device (100) of the fault prediction system of a smart factory according to the present embodiment further includes a sound detection unit (150) that detects sound generated during the operation of the automation equipment (10), and a sound inspection unit (160) that converts the sound signal detected by the sound detection unit (150) into sound data and detects abnormal sounds in the sound data.

[0040] The control unit (170) processes the sound data of the sound inspection unit (160).

[0041] The control unit (170) of the fault prediction system of the smart factory according to the present embodiment generates a timestamp of the time of occurrence of a defect determined by the vision inspection unit (120), generates defect indication data by matching vibration data and sound data at the corresponding time, and stores and manages the same.

[0042] The above defect indication data is characterized by including vision data of a defective product, vibration data of the time of the defect determination, and sound data of the time of the defect determination.

[0043] The artificial intelligence can be trained by accumulating the above defect indicator data.

[0044] The detection device (100) of the fault prediction system of a smart factory according to the present embodiment further includes a data preprocessing module (180) and a detection communication module (190).

[0045] The above data preprocessing module (180) removes noise unnecessary for learning from accumulated defect indicator data and standardizes the data scaling.

[0046] In this embodiment, the product (20) may be an injection molded product, a forged product, a machined product, etc.

[0047] The fault prediction system of a smart factory according to the present embodiment is characterized by determining a fault indication of an automated facility (10) by combining at least one of vibration or sound at the time of production of a defective product.

[0048] Defective products can be caused by various variables such as temperature, humidity, raw material defects, and mold defects. The fault prediction system of the smart factory according to the present embodiment is characterized by matching the timestamp at the time when a defective product is produced (not the cause) with the output value of the automated equipment (10) (vibration and sound in the present embodiment).

[0049] The fault prediction server (30) includes a server communication module (210) and an AI learning engine (220) that learns the vibration or sound when a defective product is produced by learning the accumulated fault indication data.

[0050] The AI ​​learning engine (220) optimizes the failure prediction algorithm by utilizing the stored failure indicator data as a training dataset.

[0051] The fault prediction system of a smart factory according to the present embodiment further includes a comparison judgment module (250) that determines signs of failure through a fault prediction algorithm learned through the AI ​​learning engine (220).

[0052] In this embodiment, the comparison judgment module (250) is placed in the detection device (100) for real-time judgment. Unlike this embodiment, the comparison judgment module (250) may be placed in the fault prediction server (30).

[0053] The vision detection unit (110) is a component that generates vision data by capturing images of products (20) produced in the automation equipment (10) in real time. In order for the fault prediction system of the present invention to maintain high precision, an organic combination of an optical device and related modules capable of capturing even minute defects in the products is essential.

[0054] The above vision detection unit (110) uses an industrial camera equipped with an image sensor that converts the appearance of the product (20) into a digital image signal.

[0055] A camera equipped with a high-speed global shutter function can be used to capture images without ghosting at the speed of the production line.

[0056] Depending on the shape or inspection range of the product (20), a telecentric lens can be used to enable precise dimensional measurement by minimizing distortion.

[0057] An industrial LED lighting module that highlights defects by controlling the reflection of light according to the material of the product (20) (injection molding, forging, machining, etc.) can be used.

[0058] It includes a dome light, ring light, or bar light that is selectively applied depending on the surface condition of the object to be inspected.

[0059] The above vision detection unit (110) further includes a photo sensor or a laser distance sensor that detects the entry of a product (20) on a conveyor of the automation equipment (10) and determines the shooting time of the camera. It may further include a strobe controller that receives a trigger signal collected from the sensor and synchronizes the camera and lighting in nanosecond (ns) units.

[0060] In the fault prediction system of a smart factory according to the present embodiment, the vibration detection unit (130) is configured by adopting a piezoelectric vibration detection sensor to precisely monitor the mechanical driving state of the automation equipment (10).

[0061] Piezoelectric vibration sensors utilize the piezoelectric effect, which generates electrical charge when physical pressure or deformation is applied to piezoelectric elements such as quartz or ceramics. They possess the characteristics of self-generating sensors capable of directly converting vibration energy into electrical signals without the need for a separate external power supply.

[0062] The above vibration detection unit (130) includes an internal piezoelectric element and an inertial mass that applies a load to the element, and when vibration occurs in the automation equipment (10), the mass applies a force proportional to the acceleration to the piezoelectric element, thereby outputting a minute charge corresponding to the magnitude and frequency of the vibration.

[0063] These piezoelectric sensors have excellent durability due to their structural absence of moving parts, and enable stable data extraction even in extreme smart factory environments where high temperatures, high pressures, and repetitive shocks occur, such as in the production processes of injection molded or forged products.

[0064] In addition, the frequency response range is very wide, so there is an advantage in that it can detect low-frequency rotational vibrations of the drive motor of the automation equipment (10) as well as fine high-frequency friction sounds caused by wear of bearings or gears over a wide range.

[0065] The fine charge signal generated in the vibration detection unit (130) is transmitted to the vibration inspection unit (140), amplified into a voltage signal through a charge amplifier provided inside the vibration inspection unit (140), and then converted into digital data and acquired.

[0066] The control unit (170) generates a timestamp at the time when a defect in the product (20) is determined by the vision inspection unit (120), and generates defect indication data by extracting vibration data based on the piezoelectric sensor synchronized with the timestamp.

[0067] The above vibration detection unit (130) is firmly attached to a major vibration point, such as a motor housing or bearing block of the automation equipment (10), using a stud connection method to maximize vibration transmission efficiency and minimize signal loss.

[0068] High-sensitivity vibration data collected through a piezoelectric sensor is used as a core input parameter of the AI ​​learning engine (220) along with sound data, and through this, the system performs precise failure prediction by pattern analysis beyond simple vibration numerical comparison.

[0069] In the fault prediction system of a smart factory according to the present embodiment, the sound detection unit (150) is a component that detects an abnormal state of the equipment by precisely collecting acoustic signals in the audible frequency and inaudible frequency bands that occur during the operation of the automated equipment (10).

[0070] The above sound detection unit (150) is configured to include a high-sensitivity microphone module to collect specific noises generated by mechanical friction, impact, air leakage, etc. of the automation equipment (10).

[0071] The above microphone module is designed to provide high mounting capability even in confined installation spaces and maintain constant sensitivity even in harsh industrial environments by adopting a MEMS (Micro-Electro Mechanical Systems) microphone utilizing ultra-small precision mechanical technology.

[0072] Additionally, the sound detection unit (150) may include a directional microphone that intensively collects sound from a specific direction to separate and collect only the unique operating sound of the equipment from surrounding machine operating sounds or background noise, or a microphone array in which a plurality of microphones are arranged.

[0073] The above microphone array applies beamforming technology to amplify noise signals generated at specific locations and cancel out unnecessary ambient noise, thereby enhancing the purity of sound data directly related to signs of failure.

[0074] The sound detection unit (150) may be configured with an integrated pre-processing circuit and pre-amplifier that suppress noise and correct the signal before converting the collected analog acoustic signal into a digital signal.

[0075] The sound signal detected through the sound detection unit (150) is transmitted to the sound inspection unit (160) and converted into sound data in the time and frequency domains, and the control unit (170) synchronizes this with the defect judgment timestamp of the vision inspection unit (120) to generate defect indication data.

[0076] In particular, the sound detection unit (150) includes a sensor capable of detecting not only the audible range but also the ultrasonic range (Ultrasound) which is inaudible to humans, thereby providing a technical advantage of being able to detect early defects in the bearing or minute pressure vessel leakage before vision or vibration data.

[0077] The above sound detection unit (150) is positioned near the motor, gearbox, or cylinder, which are the main sources of noise in the automation equipment (10), and is installed on a vibration-damping structure or shock mount to prevent the vibration of the equipment from being directly transmitted to the microphone and causing noise.

[0078] The configuration of the sound detection unit (150) enables auditory analysis of abnormal behavior inside the equipment that is difficult to determine solely from vibration data, and provides core basic data that allows the AI ​​learning engine (220) to build a more three-dimensional fault prediction model.

[0079] In the fault prediction system of a smart factory according to the present embodiment, the data preprocessing module (180) performs the role of processing raw data collected from the detection device (100) into a form suitable for AI learning and real-time comparative analysis.

[0080] The data preprocessing module (180) receives data generated by the vision inspection unit (120), the vibration inspection unit (140), and the sound inspection unit (160), and removes factors that reduce the accuracy of the fault prediction.

[0081] First, the data preprocessing module (180) performs filtering processing to remove unnecessary signals, such as electrical noise or environmental noise from an industrial site, included in the collected data.

[0082] For vibration and sound data, a bandpass filter or a moving average filter that allows only specific frequency bands to pass is applied to minimize signal distortion and preserve only valid feature points.

[0083] In addition, the data preprocessing module (180) performs data scaling so that vision, vibration, and sound data having different physical quantities and unit ranges can be compared on the same computational line.

[0084] The above normalization includes normalization, which converts the range of data to between 0 and 1, or standardization, which converts the mean to 0 and the standard deviation to 1, thereby preventing the AI ​​learning results from being biased by the magnitude of specific data values.

[0085] The data preprocessing module (180) performs time synchronization processing to align the time axis of each sensor data based on the timestamp generated by the control unit (170).

[0086] This is intended to ensure the reliability of defect indication data by aligning the vibration and sound data from the time when a defect is detected by vision data and the section immediately preceding it so that they physically point to the same process moment.

[0087] The above module may include a feature extraction function that reduces the dimensionality of data by extracting only meaningful features to efficiently process high-volume data.

[0088] For example, kurtosis, skewness, or energy density at specific frequencies are calculated from vibration signals to convert vast time series data into a concise feature vector form.

[0089] The operation method of this data preprocessing module (180) provides the effect of maximizing the learning efficiency and judgment accuracy of the fault prediction algorithm by removing noise unnecessary for learning from accumulated fault indicator data and supplying only refined data to the AI ​​learning engine (220).

[0090] Consequently, the data preprocessing module (180) serves as a key gateway for converting rough processing data from the field into high-quality information assets for intelligent fault prediction.

[0092] In the fault prediction system of a smart factory according to the present embodiment, the AI ​​learning engine (220) is provided within the fault prediction server (30) and is a core computational module that analyzes fault indication data transmitted from the detection device (100) and generates an optimized fault prediction algorithm.

[0093] The above AI learning engine (220) learns the complex correlation between the collected vision, vibration, and sound data to pattern minute signs of equipment abnormalities that are difficult to identify by visual inspection or simple numerical comparison.

[0094] First, the AI ​​learning engine (220) receives the defective symptom data refined through the data preprocessing module (180) and constructs a training dataset.

[0095] The above learning dataset is managed by combining the image feature value at the time of defect occurrence determined by the vision inspection unit (120) and the feature vector of vibration data and sound data synchronized at that time into a single label.

[0096] The specific operation method of the AI ​​learning engine (220) involves a multidimensional feature extraction step, and the peak component in the frequency domain is extracted from vibration data, and the time-series energy change on the spectrogram is extracted from sound data to quantify the state of the equipment.

[0097] The learning engine adopts a Convolutional Neural Network (CNN) algorithm to learn the spatial features of imaged sound and vibration patterns, and analyzes failure trends over time based on the equipment status using a Long Short-Term Memory (LSTM) algorithm.

[0098] The above AI learning engine (220) establishes a unique data distribution that appears when a failure threshold is reached as a failure indicator model by learning by comparing the data pattern during normal operation with the failure indicator data pattern immediately before a failure occurs.

[0099] The fault prediction algorithm that has been trained is loaded into a comparison judgment module (250), and the comparison judgment module (250) calculates the cosine similarity between the detection data flowing in in real time and the trained model and constantly monitors whether the similarity is less than 90%.

[0100] The AI ​​learning engine (220) has a self-evolving characteristic of correcting judgment errors caused by changes in the process environment or equipment aging by performing a re-training process that updates the existing model whenever new defect indicator data occurs.

[0101] In addition, the AI ​​learning engine (220) does not stop at simply determining whether there is a failure, but classifies what type of failure (e.g., bearing damage, lack of lubrication, motor overload, etc.) the current state of the equipment is close to based on the learned data and provides detailed information to the manager.

[0102] The operating method of this AI learning engine (220) provides the effect of dramatically improving the effectiveness and precision of failure prediction by learning a clear correlation, such as the ‘equipment output value at the time of product failure,’ even among complex process variables where it is difficult to identify the cause.

[0103] A plurality of the above-mentioned automation equipment (10) may be arranged and connected via a conveyor. The automation equipment (10) may perform a process for producing a product. The above-mentioned automation equipment (10) may include a robot arm capable of moving the product.

[0104] The products processed or produced through the above automation equipment (10) may include intermediate products.

[0105] The above automation equipment (10) includes a cabinet (300).

[0106] The cabinet (300) includes an entrance (301) and an exit (302), and the remaining sides excluding the entrance (301) and the exit (302) are closed.

[0107] The above sound inspection unit (160) is installed inside the cabinet (300).

[0108] The cabinet (300) is formed in the shape of a rectangular prism overall.

[0109] The cabinet (300) comprises a plurality of vertically arranged columns (310), a bottom panel (320) connecting the columns (310) to form a bottom surface, a top panel (330) connected to the top of the columns (310) to form a top surface, and a side panel (340) connecting the columns (310) to form a side surface.

[0110] The above column (310) includes a first column (311), a second column (312), a third column (313), and a fourth column (314).

[0111] The above vibration detection unit (130) is placed in at least one of the first column (311), the second column (312), the third column (313), or the fourth column (314).

[0112] In this embodiment, the vibration detection unit (130) is installed at the top of the column (310).

[0113] Based on the first column (311), the second column (312), the third column (313), and the fourth column (314) are arranged in a clockwise or counterclockwise direction.

[0114] The above side panel (340) includes a first side panel (341), a second side panel (342), a third side panel (343), and a fourth side panel (344).

[0115] An inlet (301) is positioned in any one of the first side panel (341), second side panel (342), third side panel (343), or fourth side panel (344), and an outlet (302) is positioned in any one of the others.

[0116] In this embodiment, the inlet (301) and the outlet (302) form an angle of 90 degrees when viewed from the top. The vision inspection unit (120) is positioned on the outlet (302) side to inspect the product (20).

[0117] The above vibration detection unit (130) includes a vibration detection module (400) and a module mount (490) that installs the vibration detection module (400) apart from the column (310).

[0118] The above vibration sensing module (400) comprises a base (410) which is assembled with the module mount (490) and spaced upward from the top of the column (310), a post (420) which is installed vertically on the base (410), an inertial mass (430) which is arranged to surround the post (420), a piezoelectric element (440) which is arranged between the post (420) and the inertial mass (430), a preload ring (450) which is arranged to surround the inertial mass (430), and an ICP circuit (470) which converts and amplifies a high-impedance charge signal output from the piezoelectric element (440) into a low-impedance voltage signal.

[0119] The vibration sensing module (400) according to the present embodiment is configured to include a base (410), an acoustic shield (415), a post (420), an inertial mass (430), a piezoelectric element (440), a preloading ring (450), and an ICP circuit (460) to maximize the piezoelectric effect and block external noise.

[0120] The base (410) is directly connected to the housing or bearing block of the automation equipment (10) and serves as a foundation support that transmits vibration energy generated from the equipment into the module.

[0121] The acoustic shield (415) is combined with the base (410) and is provided in the form of a housing that encloses the internal components, and blocks acoustic noise or external electromagnetic waves transmitted into the air so as not to interfere with vibration sensing, thereby helping to purely collect only the vibration signals transmitted through the facility structure.

[0122] The post (420) is installed extending vertically from the upper center of the base (410) and serves as a central axis that supports and fixes the piezoelectric element (440) and the inertial mass (430), thereby preventing movement of the internal components due to external vibrations.

[0123] The inertial mass (430) is formed in a shape that surrounds the post (420) and is positioned to surround the piezoelectric element (440). When vibration acceleration occurs due to the operation of the equipment, physical compression or tension is applied to the lower piezoelectric element (440) by attempting to maintain its original position due to inertia.

[0124] The piezoelectric element (440) is interposed between the post (420) and the inertial mass (430) and generates an electric charge in response to a physical force transmitted from the inertial mass (430), and outputs a minute charge proportional to the magnitude of the acceleration in real time to convert mechanical vibration into an electrical signal.

[0125] The preload ring (450) is installed to apply a constant preload to the piezoelectric element (440) and the inertial mass (430), and maintains a close contact between the components even when the magnitude of the vibration is small or the direction changes rapidly, thereby ensuring signal linearity and increasing measurement precision.

[0126] The ICP circuit (460) is an integrated circuit that converts and amplifies a high-impedance charge signal output from the piezoelectric element (440) into a low-impedance voltage signal, and transmits refined vibration data to the vibration inspection unit (140) by minimizing noise interference that may occur during the transmission process.

[0127] The above module mount (490) includes a mount column (495) whose upper end is coupled to the lower surface of the base (410) and whose lower end is coupled to the upper end of the column (310) and fixed, a mount screw thread (496) formed on the outer surface of the mount column (495), an upper fixing member (491) which is screw-assembled with the mount column (495) and is in close contact with the lower surface of the base (410) to fix the mount column (495) and the base (410), and a lower fixing member (492) which is screw-assembled with the mount column (495) and is in close contact with the upper surface of the column (310) to fix the mount column (495) and the column (310).

[0128] A column assembly groove (315) is formed concavely downward from the upper surface of the column (310), screw threads are formed on the inner circumference of the column assembly groove (315), and the lower end of the mount column (495) is screw-coupled with the column assembly groove (315).

[0129] The lower fixing member (492) is screw-assembled with the mount column (495) and is pressed against the top of the column (310), preventing the screw connection between the mount column (495) and the column assembly groove (315) from becoming loose or unfastened.

[0130] A base assembly groove (412) is formed concavely upward from the bottom surface of the base (410), and the upper end of the mount column (495) is fixed by screw coupling with the base assembly groove (412).

[0131] The upper fixing member (491) is screw-assembled with the mount column (495) and is in close contact with the bottom surface of the base (410), preventing the screw connection between the mount column (495) and the base assembly groove (412) from becoming loose or unraveled.

[0132] Since the vibration detection module (400) is fixed with the upper and lower ends of the mount column (495) fixed and spaced apart from the column (310), the vibration transmitted to the cabinet (300) can be transmitted more effectively to the interior, and thereby, abnormal vibration can be effectively detected.

[0133] A control method for a fault prediction system of a smart factory according to an embodiment of the present invention comprises: a data acquisition step (S10) in which multidimensional process data is collected in real time through a vision detection unit (110), a vibration detection unit (130), and a sound detection unit (150) of a detection device (100) simultaneously with the operation of an automation facility (10); a vision inspection step (S20) in which the vision inspection unit (120) analyzes the vision data and determines whether the product (20) is good or bad; a data integration step (S30) in which, when a defective product is determined in the vision inspection step (S20), a control unit (170) generates a timestamp at that time and matches the vibration data and sound data synchronized with it to generate defect indication data; a data preprocessing step (S40) in which the data preprocessing module (180) removes noise from the defect indication data, unifies the data specifications, and transmits them to a fault prediction server (30); and the fault prediction server (30). The system may be configured to include an AI learning step (S50) in which an AI learning engine (220) learns the failure pattern of the equipment based on accumulated failure indicator data and optimizes the failure prediction algorithm, a similarity analysis step (S60) in which the comparison judgment module (250) compares real-time incoming vibration and sound data with the learned failure prediction algorithm to calculate similarity, and a failure prediction notification step (S70) in which, if the similarity calculated in the similarity analysis step (S60) is less than 90%, it is determined to be a failure indicator of the automated equipment (10) before actual failure occurs and outputs a preventive maintenance notification.

[0134] In the fault prediction system of a smart factory according to the present embodiment, the data acquisition step (S10) is a base step of collecting physical state information of the equipment and quality information of the product in real time through multiple sensors in the detection device (100) simultaneously with the operation of the automated equipment (10).

[0135] The above data acquisition step (S10) is performed continuously from the starting point of the production line to the exit (302), and the collected data includes vision data, vibration data, and sound data.

[0136] Specifically, when the automation equipment (10) is operated, a product (20) is fed through the entrance (301) of the cabinet (300), and the drive parts of the equipment, including a robot arm, perform injection, forging, or cutting processes.

[0137] At this time, the vibration detection unit (130) detects fine vibrations generated in the equipment by using a vibration detection module (400) installed via a module mount (490) on the top of the column (310) of the cabinet (300).

[0138] The piezoelectric element (440) in the vibration detection module (400) converts the physical pressure applied by the inertial mass (430) into a charge signal, and the ICP circuit (460) amplifies this into a low-impedance voltage signal to generate real-time vibration data.

[0139] At the same time, a sound detection unit (150) installed inside the cabinet (300) detects noise generated during equipment operation and acquires pure operation sound data with ambient noise suppressed through a directional microphone or a MEMS microphone array.

[0140] The acquired vibration and sound data is temporarily stored in a time-series manner in the ring buffer system of the control unit (170) and subsequently maintains a waiting state to be matched with the vision inspection time.

[0141] Meanwhile, the vision detection unit (110) captures an image of the product (20) in response to a trigger signal from the photo sensor when the product (20) moving along the conveyor reaches an inspection area near the exit (302).

[0142] The above-mentioned captured image is transmitted to the vision inspection unit (120) as vision data, and all these data acquisition processes are organically interconnected to become a source of big data for fault prediction.

[0143] This data acquisition step (S10) is characterized by collecting vibrations transmitted to the cabinet (300) with higher purity by fixing the vibration detection module (400) apart from the column (310) through the module mount (490), thereby increasing the reliability of abnormal sign detection.

[0144] In the fault prediction system of a smart factory according to the present embodiment, the vision inspection step (S20) is a process in which a vision inspection unit (120) positioned at the exit (302) side of the automation equipment (10) analyzes the external quality of the product (20) in real time and generates a defect judgment signal that serves as a reference point for fault prediction.

[0145] The vision inspection unit (120) first performs a preprocessing process to remove noise and highlight feature points from high-resolution image data obtained through the industrial camera of the vision detection unit (110).

[0146] In the subsequent analysis process, the vision inspection unit (120) compares the currently extracted vision data with the master data of a pre-trained artificial intelligence algorithm or a pre-set normal product.

[0147] When the above-mentioned product (20) is an injection molded product, a forged product, or a machined product, the vision inspection unit (120) precisely inspects defect items such as surface scratches, cracks, under-dimensions, and incomplete molding.

[0148] As a result of the inspection, if the defect value of the product (20) exceeds a preset threshold and is determined to be a defective product, the vision inspection unit (120) immediately sends a defect determination signal to the control unit (170).

[0149] At this time, the vision inspection unit (120) does not stop at simply performing a pass / fail judgment, but also preserves the characteristic value of the vision data that serves as the basis for the judgment of defect, along with a timestamp.

[0150] The point in time at which a defect is determined in the above vision inspection step (S20) becomes the synchronization standard for the entire system, and the control unit (170) specifically extracts sound data inside the cabinet (300) and vibration data at the top of the column (310) based on this point in time.

[0151] That is, the vision inspection step (S20) is characterized by detecting product quality abnormalities and simultaneously acting as a trigger to capture physical abnormal signs (vibration and sound) that may have occurred inside the equipment as defect indicator data.

[0152] Through this organic connection, the system provides a learning basis for distinguishing whether the cause of the defect is due to external variables such as temperature, humidity, and raw materials, or due to mechanical defects of the automation equipment (10) itself.

[0153] In the fault prediction system of a smart factory according to the present embodiment, the data integration step (S30) is a process of generating defect indication data, which is a core asset of AI learning, by aligning quality events detected by the vision inspection unit (120) and physical signals collected from vibration and sound sensors along the time axis.

[0154] The above data integration step (S30) is executed by the control unit (170) at the moment when a specific product (20) is determined to be defective in the vision inspection step (S20).

[0155] When the control unit (170) receives a defect judgment signal from the vision inspection unit (120), it immediately generates a timestamp at that time and searches for vibration data and sound data that were temporarily stored in the ring buffer in the data acquisition step (S10) based on the timestamp.

[0156] At this time, the control unit (170) specifically extracts data for a predetermined past time interval (e.g., 5 to 10 seconds) from the timestamp point in order to secure a signal generated at a process step that is physically identical to the time of defect determination.

[0157] The extracted data is integrated into defect indication data by combining the feature values ​​of vision data determined to be defective, the vibration data at the corresponding point in time, and the sound data into a single set.

[0158] The data integration step (S30) according to the present embodiment is technically characterized by matching the equipment output values ​​(vibration and sound) at the time when the defective product is produced, rather than determining whether the cause of the defect is due to external variables such as temperature, humidity, raw material defects, or mold defects.

[0159] This matching method enables the isolation of equipment-specific abnormal behavior patterns from complex environmental variables, thereby simplifying the causal relationships of failure prediction and improving accuracy.

[0160] The fault indication data generated by the control unit (170) is temporarily stored in memory within the detection device (100) or transmitted to the fault prediction server (30) via the detection communication module (190) and accumulated and managed in the database.

[0161] The above accumulated defect indicator data is subsequently used as a learning dataset for the AI ​​learning engine (220) and serves as a basis for establishing the characteristic vibration and acoustic indicators that appear just before the equipment actually fails.

[0162] Consequently, the data integration step (S30) performs a core process of converting a simple list of data into knowledge information for intelligent fault prediction by organically combining heterogeneous multi-sensor data centered on quality events.

[0163] In the fault prediction system of a smart factory according to the present embodiment, the data preprocessing step (S40) is a process of refining heterogeneous and vast amounts of raw data collected from a detection device (100) into a high-quality dataset suitable for artificial intelligence learning and real-time comparison judgment.

[0164] The data preprocessing module (180) receives data generated by the vision inspection unit (120), the vibration inspection unit (140), and the sound inspection unit (160), and performs processing to remove noise that lowers the reliability of the data and to unify the specifications of different physical quantities.

[0165] The above data preprocessing step (S40) first undergoes a digital filtering step to remove unnecessary components, such as electrical interference signals or mechanical environmental noise from the smart factory site, included in the collected data.

[0166] For vibration and sound data, a bandpass filter that allows only specific frequency bands to pass is applied to block low-frequency vibrations or high-frequency noise unrelated to equipment failure, thereby enhancing the clarity of abnormal sign patterns.

[0167] In addition, the data preprocessing module (180) performs data scaling to prevent learning bias that may occur due to differences in the units and numerical ranges of vision, vibration, and sound data.

[0168] The above standardization includes normalization, which converts the data into a range between 0 and 1 using the minimum and maximum values ​​of the data, or standardization, which readjusts the distribution using the mean and standard deviation, thereby enabling the AI ​​learning engine (220) to analyze multidimensional data on the same weight line.

[0169] In particular, in the data preprocessing step (S40), time synchronization alignment processing is performed to finely adjust and match the time axes of vision, vibration, and sound data based on the timestamp generated by the control unit (170).

[0170] This is intended to clarify the causal relationship of defect indicator data by placing the equipment operation signal immediately preceding the moment when a defective product is determined on a process sequence that physically matches.

[0171] In addition, the above module can perform feature value optimization processing to reduce the dimensionality of the data and extract meaningful statistical indicators (e.g., RMS of vibration, kurtosis, frequency energy density of sound, etc.) in order to reduce the computational load and highlight only key features.

[0172] The precise operation method of this data preprocessing step (S40) provides a technical basis for maximizing the learning efficiency of the AI ​​learning engine (220) and improving the accuracy of the final fault indication judgment by converting unrefined field data into structured information assets for intelligent fault prediction.

[0173] In the fault prediction system of a smart factory according to the present embodiment, the AI ​​learning step (S50) is a process in which the AI ​​learning engine (220) of the fault prediction server (30) analyzes accumulated fault indicator data to diagnose the condition of the equipment and optimizes a prediction model that quantifies the probability of future failure occurrence.

[0174] The above AI learning step (S50) is performed by receiving a high-quality learning dataset refined through the data preprocessing step (S40) and learns the complex correlation between vision, vibration, and sound data in a high-dimensional feature space.

[0175] Specifically, the AI ​​learning engine (220) matches the type of defective product determined by the vision inspection unit (120) with the vibration and sound patterns at that time to learn the causal mechanism in which the mechanical abnormal behavior of the equipment leads to quality defects in the actual product.

[0176] In this embodiment, the AI ​​learning engine (220) utilizes a convolutional neural network (CNN) optimized for imaged time-series signal analysis to extract spatial features of vibration spectrum and sound spectrogram.

[0177] At the same time, by combining the Long Short-Term Memory (LSTM) algorithm to track the trend of signal changes over equipment operating time, learning is performed to recognize gradual performance degradation patterns rather than simple instantaneous outliers.

[0178] The prediction model constructed through the above AI learning step (S50) forms an optimal boundary surface that distinguishes between the data distribution of a normal operating state and the data distribution of defect indications immediately before failure.

[0179] The learning engine performs a retraining process that updates weights whenever new defect indicator data is accumulated, thereby possessing intelligent characteristics that autonomously correct judgment errors caused by changes in temperature or humidity or the aging of the equipment.

[0180] In addition, this step establishes points where energy in a specific frequency band surges or the entropy of the sound pattern changes as key indicators of fault signs, thereby enhancing the judgment reliability of the fault prediction algorithm.

[0181] The fault prediction algorithm optimized in this way is then passed to the similarity analysis step (S60) and used for comparison with real-time incoming data.

[0182] Consequently, the AI ​​learning stage (S50) plays a key role in transforming fault diagnosis, which relied on the experience of skilled workers, into a precise, data-based scientific prediction system by enhancing artificial intelligence based on a clear answer key, the 'output value at the time of defective product production'.

[0183] In the fault prediction system of a smart factory according to the present embodiment, the similarity analysis step (S60) is a process of comparing the driving signal of the equipment collected in real time with the fault prediction model built through the AI ​​learning step (S50) to quantitatively calculate how much the current state of the equipment matches the signs immediately before the occurrence of a past defect.

[0184] The similarity analysis step (S60) is performed by a comparison judgment module (250), and the comparison judgment module (250) can be placed inside the detection device (100) to perform calculations in a real-time edge computing manner or placed in the fault prediction server (30) to perform calculations in a centralized manner.

[0185] The above comparison judgment module (250) receives real-time data input through the vibration detection unit (130) and sound detection unit (150) during the operation of the automation equipment (10).

[0186] The real-time data input at this time passes through a data preprocessing module (180) to remove noise and is standardized, and is converted into a feature vector of the same dimension as the learned defect indicator data.

[0187] The comparison judgment module (250) calculates the distance between the current feature vector and the failure symptom pattern stored in the learning engine, and in this embodiment, calculates the similarity by utilizing a cosine similarity algorithm that measures the degree of directional agreement between data.

[0188] The calculated similarity value is expressed as a numerical value between 0 and 100%, and this serves as an indicator of how similar the vibration and sound patterns currently occurring in the equipment are to past abnormal equipment behaviors that caused actual product defects.

[0189] In the similarity analysis step (S60), instead of simply comparing the overall signal magnitude, the AI ​​focuses on comparing key failure factors learned, such as variability in specific frequency bands or abnormal repetition periods of sound waveforms.

[0190] The comparison judgment module (250) above constantly monitors the similarity value calculated in real time and determines whether the value falls below a preset warning threshold of 90%.

[0191] If the similarity is 90% or higher, the equipment is considered to be in a stable state, and the data acquisition step (S10) is continued while monitoring is performed.

[0192] If the similarity is calculated to be less than 90%, this means that even if no product defects are found in the vision inspection unit (120) yet, the mechanical deterioration inside the equipment has reached a failure threshold, so it immediately proceeds to the next step, the failure prediction notification step (S70).

[0193] Unlike conventional technology focused on post-action, the operation method of this similarity analysis step (S60) provides technical advantages that dramatically increase the accuracy of failure prediction and minimize false alarms by comparing a clear reference model, such as 'output value when defective products are produced,' with real-time data.

[0194] In the fault prediction system of a smart factory according to the present embodiment, the fault prediction notification step (S70) is a final decision-making process that notifies the manager of potential risks to the equipment based on the calculation result of the similarity analysis step (S60) and performs active process control based on the system's judgment to prevent major accidents and mass production of defects.

[0195] The fault prediction notification step (S70) is entered immediately when the similarity between the real-time data calculated by the comparison judgment module (250) and the fault prediction algorithm is less than 90%, which is a preset threshold value.

[0196] At this stage, the fault prediction server (30) outputs a preventive maintenance notification to an administrator terminal or field control panel (HMI) through the server communication module (210) or the detection communication module (190).

[0197] The above preventive maintenance notification may include the current similarity value of the equipment, the expected failure location (e.g., wear of the motor bearing in the first column), and recommended inspection measures as visual or auditory signals.

[0198] In addition, the fault prediction notification step (S70) of the present invention does not merely transmit a notification, but also monitors the quality status of the product (20) produced in real time through the vision detection unit (110).

[0199] When a sign of failure is detected and the real-time defect rate of the product calculated by the vision inspection unit (120) exceeds a preset standard value, the system determines that the physical threshold of the equipment has reached a limit.

[0200] Accordingly, the system may further include an equipment stop step (S80) forcibly stopping the automated equipment (10) as a sub-process or subsequent step of the fault prediction notification step (S70).

[0201] In the above equipment stop step (S80), the control unit (170) works in conjunction with the PLC (Programmable Logic Controller) of the automation equipment (10) to cut off the power to the drive motor or stop the operation of the robot arm in an emergency, thereby preventing damage to the equipment and preventing defective products from entering the subsequent process.

[0202] At this time, the vision detection unit (110) acquires and stores an image of the last product produced just before the equipment stops, and this can be used as comparison data to verify whether the equipment is operating normally after maintenance is completed.

[0203] The configuration of the fault prediction notification step (S70) and the equipment stop step (S80) combines 'prediction' by artificial intelligence and 'actual results' by vision inspection to double-secure process safety, thereby completing the function as an intelligent autonomous control system beyond a simple monitoring system.

[0205] Although specific embodiments have been described in the detailed description of the present invention, it is understood that various modifications are possible within the scope of the invention. Therefore, the scope of the present invention should not be limited to the described embodiments, but should be defined by the claims set forth below as well as equivalents thereof. Explanation of the symbols

[0207] 10: Automation Equipment 20: Products 30: Fault Prediction Server 100: Detector 110: Vision Sensing Unit 120: Vision Inspection Department 130: Vibration detection unit 140: Vibration Inspection Department 150: Sound detection unit 160: Sound Inspection Department 170: Control unit 180: Data Preprocessing Module 190: Sensing communication module 210: Server communication module 220: AI learning engine 250: Comparison and judgment module 300: Cabinet 301: Entrance 302: Exit 310: Column 315: Column assembly groove 320: Bottom panel 330: Top Panel 340: Side panel 400: Vibration detection module 410: Bass 412: Base assembly groove 415: Acoustic Shield 420: Post 430: Inertial mass body 440: Piezoelectric element 450: Preloading 460: ICP circuit 490: Module Mount 491: Upper fixing member 492: Lore Fixed Part 495: Mount Column 496: Mount thread

Claims

Claim 1 A fault prediction method for a fault prediction system of a smart factory, comprising a detection device (100) that is physically connected to and installed with an automation facility (10) and generates and extracts data from the automation facility (10), and a fault prediction server (30) that is connected to the detection device (100) via a wired / wireless communication network and receives and processes defect indication data, wherein the automation facility (10) includes a cabinet (300), and the detection device (100) comprises a vision detection unit (110) that captures an image of a product (20) produced in the automation facility (10), a vision inspection unit (120) that extracts vision data from the image captured by the vision detection unit (110) and determines whether the product is a defective product or a normal product through the extracted vision data, a vibration detection unit (130) that detects vibration of the automation facility (10), and a vibration detection unit (130) that converts a signal detected by the vibration detection unit (130) into vibration data and detects abnormal vibration in the vibration data The system includes a vibration inspection unit (140) that makes a judgment, a control unit (170) that is electrically connected to the vision inspection unit (120) and the vibration inspection unit (140) and processes the vision data and vibration data, and a data acquisition step (S10) that collects multidimensional process data in real time through the vision detection unit (110), vibration detection unit (130), and sound detection unit (150) of the detection device (100) simultaneously with the operation of the automation equipment (10), a vision inspection step (S20) in which the vision inspection unit (120) analyzes the vision data to determine whether the product (20) is good or bad, a data integration step (S30) in which, when a defective product is determined in the vision inspection step (S20), the control unit (170) generates a timestamp at that time and generates defect indication data by matching the vibration data and sound data synchronized with the timestamp, and the data preprocessing module (180) of the detection device (100) A data preprocessing step (S40) for removing noise from fault indicator data and standardizing data specifications and transmitting them to a fault prediction server (30), andThe AI ​​learning engine (220) of the fault prediction server (30) learns the failure pattern of the equipment based on accumulated failure indicator data and optimizes the fault prediction algorithm in an AI learning step (S50); the comparison judgment module (250) compares vibration and sound data flowing in in real time with the learned fault prediction algorithm to calculate a similarity in a similarity analysis step (S60); and if the similarity calculated in the similarity analysis step (S60) is less than 90%, it is determined to be a failure indicator of the automated equipment (10) prior to actual failure occurrence and outputs a preventive maintenance notification in a fault prediction notification step (S70); the vibration detection unit (130) includes a vibration detection module (400) and a module mount (490) that installs the vibration detection module (400) apart from the cabinet (300); the vibration detection module (400) is assembled with the module mount (490); and the The apparatus comprises a base (410) spaced upward from the top of the cabinet (300), a post (420) installed vertically on the base (410), an inertial mass (430) arranged to surround the post (420), a piezoelectric element (440) arranged between the post (420) and the inertial mass (430), a preloading ring (450) arranged to surround the inertial mass (430), and an ICP circuit (470) that converts and amplifies a high-impedance charge signal output from the piezoelectric element (440) into a low-impedance voltage signal. In the data acquisition step (S10), the vibration detection unit (130) detects a fine vibration generated in the equipment using a vibration detection module (400) installed via a module mount (490) on the top of the cabinet (300). The piezoelectric element (440) within the vibration detection module (400) converts the physical pressure applied by the inertial mass (430) into a charge signal, and the ICP circuit (460) amplifies the converted charge signal into a low-impedance voltage signal to generate real-time vibration data.At the same time, a sound detection unit (150) installed inside a cabinet (300) detects noise generated during equipment operation and acquires operation sound data, a fault prediction method of a smart factory fault prediction system. Claim 2 delete Claim 3 delete

Images