On-device security framework method for verifying and controlling behavioral units of intelligent autonomous systems

The on-device security framework for intelligent autonomous systems addresses the limitations of conventional security by verifying and controlling behavioral units using contextual data and risk calculations, ensuring stable operation by preventing unsafe actions.

KR103046571B1Active Publication Date: 2026-11-10SIOT
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
KR1020260025231
Authority / Receiving Office
KR · KR
Patent Type
Patents
Current Assignee / Owner
Filing Date
2026-02-09
Publication Date
2026-11-10
Estimated Expiration
2046-02-09

AI Technical Summary

Technical Problem

Conventional security technologies for intelligent autonomous systems fail to verify the safety of operational judgments and execution commands in real time, particularly in the face of internal errors, external attacks, or environmental disturbances, leading to potential accidents.

Method used

An on-device security framework that verifies and controls behavioral units by generating context data from multiple sensors, calculating risk levels using a neural network processing unit, and applying control measures such as restriction or blocking actions based on predefined security policies and environmental data.

Benefits of technology

Enables proactive safety judgments considering environmental information and temporal changes, reducing the risk of accidents by identifying and controlling internal errors and external intrusions before execution, thereby stabilizing the operation of intelligent autonomous systems.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 112026016554687-PAT00001_ABST
    Figure 112026016554687-PAT00001_ABST
Patent Text Reader

Abstract

The present invention relates to an on-device security framework method for verifying and controlling behavioral units of an intelligent autonomous system, comprising: a user terminal including a display unit and a user input unit; and a server connected to the user terminal via a wired or wireless network; wherein the server includes a communication unit; a storage unit; and a control unit; wherein the control unit is implemented to control the overall operation of the on-device security framework method, and the on-device security framework method comprises the steps of: receiving execution commands and operation requests through the communication unit before they are generated from the decision unit of the intelligent autonomous system and transmitted to the physical execution unit; identifying whether the entity that generated the operation request corresponds to an internal software process, an external input signal, or a remote control request, and then repeatedly verifying the validity of the identity and authority at each operation request time by comparing it with identity information and authority information for each entity already stored in the storage unit. A step of generating first context data including the system location, movement speed, time information, presence of surrounding workers, environmental data collected from multiple sensors, and the temporal change amount of each sensor data at the time the above operation request occurs, and identifying second context data representing the state of the current execution environment by sorting the first context data by component and combining them according to mutual correlations; a step of calculating a risk level corresponding to the operation request using a neural network processing unit (NPU) with the identity and authorization verification result and the second context data as input values, wherein the risk level is calculated to include a value calculated stepwise by comparing it with allowable range data included in predefined security policies and physical safety regulations;The method is characterized by comprising: a step of controlling to transmit the operation request to the execution unit when the calculated risk level is less than or equal to a first threshold value, controlling to perform at least one restriction control among operation speed reduction, output restriction, or path change when the risk level exceeds the first threshold value and is less than or equal to a second threshold value, and controlling to block the transmission of the operation to the execution unit or to suspend the operation until an external approval input is received when the risk level exceeds the second threshold value; and a step of controlling to generate log data in an immutable form, including the time of receipt of the operation request, identity and authorization verification results, the composition of the second context data, the risk level calculation results, the control execution history, and the final execution status, and to sequentially store them in the storage unit.
Need to check novelty before this filing date? Find Prior Art

Description

Technology Field

[0001] The present invention relates to an on-device security framework method for verifying and controlling behavioral units of an intelligent autonomous system. Background Technology

[0002] With the recent advancements in artificial intelligence and sensor technologies, intelligent autonomous systems—such as robots, autonomous vehicles, and unmanned facilities—are rapidly becoming widespread as they make decisions and take action independently in physical environments. Since these intelligent autonomous systems directly execute physical actions, such as driving, movement, and task execution, based on various sensor information collected from the external environment and internal computational results, they possess a characteristic where errors in judgment or external attacks can immediately lead to human casualties or facility damage.

[0003] Conventional security technologies have primarily focused on communication-centric protection structures, such as network perimeter security, access control, and authentication procedures, and have had limitations in verifying the safety of individual operational judgments or execution commands generated within the system in real time. In particular, in the case of intelligent autonomous systems, malfunctions, abuse of authority, or environmental perception errors can occur even by entities that have undergone proper authentication; however, conventional technologies have failed to provide structures to identify and control these risks at the internal operational stage in advance.

[0004] In addition, even when external attacks that indirectly distort system judgments occur, such as deception of sensor data, data corruption, or environmental disturbance, conventional technology often relies on individual sensors or single judgment results, making it difficult to assess safety while considering complex environmental contexts. Consequently, there was a limitation in that it could not block dangerous behavior before the execution phase and remained in a post-incident response-centered structure that only recognized abnormalities after an accident occurred. Prior art literature

[65535] Korean Patent Publication No. 10-2716460 (Oct. 07, 2024) Korean Patent Publication No. 10-2232649 (March 22, 2021) The problem to be solved

[0005] The present invention aims to solve the aforementioned problems by verifying operation requests generated by an intelligent autonomous system at a stage prior to physical execution, thereby enabling prior confirmation of whether the autonomous system's judgment result aligns with actual environmental conditions. This allows for safety judgments that comprehensively consider various environmental information and temporal change characteristics, unlike conventional methods that relied on a single judgment result or individual sensor information.

[0006] In addition, by repeatedly verifying the subject and execution context of the request whenever an action request occurs, it is possible to identify early instances of authorization misuse or abnormal judgment flows that may occur even under normal authentication conditions; accordingly, this has the effect of controlling risky behaviors caused by internal errors or accumulated judgment deviations, as well as external intrusions, before execution.

[0007] Furthermore, by making judgments based on the interrelationships between multiple environmental information and sensor data, the impact of errors or disturbances in specific information on the overall system judgment can be mitigated. This enables the intelligent autonomous system operating in a physical environment to operate more stably even under various situational changes and is expected to have the effect of generally reducing the possibility of unexpected physical accidents. This provides an on-device security framework method for behavioral unit verification and control of an intelligent autonomous system. means of solving the problem

[0008] According to one embodiment of the present invention, an on-device security framework method for verifying and controlling behavioral units of an intelligent autonomous system comprises: a user terminal including a display unit and a user input unit; and a server connected to the user terminal via a wired or wireless network; wherein the server comprises a communication unit; a storage unit; and a control unit; wherein the control unit is implemented to control the overall operation of the on-device security framework method, and the on-device security framework method comprises the steps of: receiving execution commands and operation requests through the communication unit before they are generated from the decision unit of the intelligent autonomous system and transmitted to the physical execution unit; identifying whether the entity that generated the operation request corresponds to any one of an internal software process, an external input signal, or a remote control request, and then repeatedly verifying the validity of the identity and authority at each operation request time by comparing it with identity information and authority information for each entity already stored in the storage unit. A step of generating first context data including the system location, movement speed, time information, presence of surrounding workers, environmental data collected from multiple sensors, and the temporal change amount of each sensor data at the time the above operation request occurs, and identifying second context data representing the state of the current execution environment by sorting the first context data by component and combining them according to mutual correlations; a step of calculating a risk level corresponding to the operation request using a neural network processing unit (NPU) with the identity and authorization verification result and the second context data as input values, wherein the risk level is calculated to include a value calculated stepwise by comparing it with allowable range data included in predefined security policies and physical safety regulations;The method is characterized by comprising: a step of controlling to transmit the operation request to the execution unit when the calculated risk level is less than or equal to a first threshold value, controlling to perform at least one restriction control among operation speed reduction, output restriction, or path change when the risk level exceeds the first threshold value and is less than or equal to a second threshold value, and controlling to block the transmission of the operation to the execution unit or to suspend the operation until an external approval input is received when the risk level exceeds the second threshold value; and a step of controlling to generate log data in an immutable form, including the time of receipt of the operation request, identity and authorization verification results, the composition of the second context data, the risk level calculation results, the control execution history, and the final execution status, and to sequentially store them in the storage unit.

[0009] The control unit generates a first change amount data by calculating a time-unit change amount for each type of sensor for a plurality of sensor data constituting the second context data, generates a second change amount judgment data by cross-comparing the first change amount data between sensors to determine whether it deviates from a change range that can be physically established simultaneously, and if a specific sensor data is identified in the second change amount judgment data as exhibiting a mutually contradictory change pattern within the same time interval when compared with other sensor data, controls the sensor data to classify it as unreliable data and gradually reduce its weight or exclude it from the calculation target when calculating the risk, and controls the risk to be recalculated using only the adjusted sensor data, thereby preventing an operation request generated by sensor deception, data contamination, or environmental disturbance from being transmitted to the physical execution stage.

[0010] The control unit generates first control data by decomposing the control parameters included in the operation request by component, and compares the first control data with pre-set physical guardrail data item by item, wherein the physical guardrail data is configured to include at least one limit value among a maximum rotation angle, maximum acceleration, allowable travel range, and continuous operation duration, and if any of the first control data is identified as exceeding the limit value of the physical guardrail data, the entire operation request is invalidated, or only the control parameters corresponding to the excess item are modified to generate reconstructed second control data and then re-entered into the risk calculation step, thereby controlling such that at least one of the abnormal or excessive control commands generated by the decision unit and external input is blocked before being transmitted to the execution unit.

[0011] The above control unit generates first behavioral history data by accumulating execution history, risk calculation results, and actual control execution results over a past period for operation requests repeatedly received from the same entity, and identifies behavioral patterns by entity based on the frequency of operation requests, average risk values, and restriction or blocking occurrence rates from the first behavioral history data. If it is determined that the behavioral pattern deviates from the pattern range typically allowed within the authority of the entity, the control unit controls subsequent operation requests generated by the entity by lowering the reference threshold value when calculating the risk or expanding the risk range to which restriction control is applied, thereby controlling the possibility of authority misuse or insider attacks by an entity that has undergone normal authentication to be suppressed in stages.

[0012] The control unit sorts the log generation target data, including the reception time of the operation request, identity and authorization verification results, the composition of the second context data, risk calculation results, and control execution results, in chronological order, and then generates first log feature data by calculating waveform-based feature values ​​from the log generation target data at each log generation time point, wherein the waveform-based feature values ​​are configured to include an amplitude value corresponding to the amount of data change between adjacent logs and a period value corresponding to the interval between log generation times after mapping each item of the log generation target data to a continuous signal on a time axis, and are calculated to form a unique waveform pattern for each log time point by combining the amplitude value and the period value, and the control unit controls the first log feature data by continuously combining it with the first log feature data generated at the previous log time point to form a cumulative waveform sequence, and identifies whether the log data has changed based on whether the continuity of the cumulative waveform sequence is damaged or whether a discontinuous phase change occurs in a specific section, and stores the waveform linkage information generated based on the cumulative waveform sequence and the original log data accordingly, wherein the waveform linkage information is used when calculating the waveform-based feature values ​​of the log generated thereafter By recording in a storage unit so as to be reused as a reference value, the waveform continuity of subsequently generated logs is not maintained if any of the past logs are modified or deleted; accordingly, the log data is controlled so that tampering is identified not by hash verification at the single log unit level, but by the physical irreversibility of the waveform continuity structure accumulated over time.

[0013] The control unit generates third state data by collecting system state change data, including the system internal computational load, power consumption change amount, processing delay time, and memory access frequency at the time of log generation, at each time the log data is generated; arranges the third state data in chronological order and calculates an irreversible state transition index based on the direction and magnitude of state change between adjacent log generation times, wherein the irreversible state transition index is configured so that it cannot be reproduced identically even if the log data is identical if the system physical state at the time of log generation is different; the control unit records the irreversible state transition index in a storage unit by correlating it with the log data, and subsequently controls the system to identify whether the log generation time and continuity are maintained by comparing the state transition index recalculated from the current system state with the previously stored state transition index at the time of log verification, based on whether both indicate the same transition path; accordingly, even if past log data is duplicated, moved, or modified, the same state transition index is not reconstructed unless the system physical state transition characteristics at the time of log generation are reproduced at the current time, thereby structurally excluding post-hoc falsification or re-insertion of log data. It is characterized by the fact that. Effects of the invention

[0014] By verifying action requests generated by an intelligent autonomous system at a stage prior to physical execution, it is possible to check in advance whether the autonomous system's judgment results align with actual environmental conditions. This enables safety judgments that comprehensively consider various environmental information and temporal change characteristics, unlike conventional methods that relied on a single judgment result or individual sensor information.

[0015] In addition, by repeatedly verifying the subject and execution context of the request whenever an action request occurs, it is possible to identify early instances of authorization misuse or abnormal judgment flows that may occur even under normal authentication conditions; accordingly, this has the effect of controlling risky behaviors caused by internal errors or accumulated judgment deviations, as well as external intrusions, before execution.

[0016] In addition, by making a judgment based on the interrelationship between multiple environmental information and sensor data, the impact of errors or disturbances in specific information on the overall system judgment can be mitigated. This enables the intelligent autonomous system operating in a physical environment to operate more stably even under various situational changes and is expected to have the effect of generally reducing the possibility of unexpected physical accidents. Thus, an on-device security framework method for behavioral unit verification and control of an intelligent autonomous system can be provided. Brief explanation of the drawing

[0017] FIG. 1 is a diagram illustrating the schematic configuration of an on-device security framework method for verifying and controlling behavioral units of an intelligent autonomous system according to one embodiment of the present invention. FIG. 2 is a diagram illustrating the specific configuration of an on-device security framework method for verifying and controlling behavioral units of an intelligent autonomous system according to an embodiment of the present invention. FIGS. 3 to 5 illustrate an example of operation of an on-device security framework method for verifying and controlling behavioral units of an intelligent autonomous system according to an embodiment of the present invention. Specific details for implementing the invention

[0018] Embodiments of the present invention will be described in detail below with reference to the accompanying drawings. In the drawings, identical reference numbers or symbols refer to components that perform substantially the same function, and the size of each component in the drawings may be exaggerated for clarity and convenience of explanation. However, the technical concept of the present invention and its core components and operations are not limited only to the components or operations described in the following embodiments. In describing the present invention, if it is determined that a detailed description of known technologies or components related to the present invention may unnecessarily obscure the essence of the present invention, such detailed description will be omitted.

[0019] In embodiments of the present invention, terms including ordinal numbers, such as first, second, etc., are used solely for the purpose of distinguishing one component from another, and singular expressions include plural expressions unless the context clearly indicates otherwise. Furthermore, in embodiments of the present invention, terms such as 'composed of,' 'include,' 'have,' etc., should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof. Additionally, in embodiments of the present invention, 'module' or 'part' performs at least one function or operation and may be implemented in hardware or software, or as a combination of hardware and software, or may be integrated into at least one module and implemented as at least one processor. Furthermore, in embodiments of the present invention, 'at least one' among a plurality of elements refers not only to all of the plurality of elements but also to each individual element excluding the remainder or all combinations thereof. Additionally, "configured to" may be used interchangeably with, depending on the context, for example, "suitable for," "having the capacity to," "designed to," "adapted to," "made to," or "capable of." "Configured to" does not necessarily mean that it is "specifically designed to" in hardware. Instead, in some situations, the expression "device configured to" may mean that the device is "capable of" doing so in conjunction with other devices or components.For example, the phrase “a processor configured (or set) to perform A, B, and C” may mean a dedicated processor for performing said operations (e.g., an embedded processor), or a generic-purpose processor (e.g., a CPU or an application processor) capable of performing said operations by executing one or more software programs stored in a memory device.

[0020] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the drawings. This description is intended to be detailed enough for a person skilled in the art to easily practice the invention, and it should be noted that the technical scope and concept of the present invention are not limited thereby.

[0021] FIG. 1 is a diagram illustrating the schematic configuration of an on-device security framework method for verifying and controlling behavioral units of an intelligent autonomous system according to an embodiment of the present invention, FIG. 2 is a diagram illustrating the specific configuration of an on-device security framework method for verifying and controlling behavioral units of an intelligent autonomous system according to an embodiment of the present invention, and FIG. 3 to 5 are diagrams illustrating an example of the operation of an on-device security framework method for verifying and controlling behavioral units of an intelligent autonomous system according to an embodiment of the present invention.

[0022] Referring to FIGS. 1 to 5, an on-device security framework method for verifying and controlling behavioral units of an intelligent autonomous system according to one embodiment of the present invention comprises: a user terminal including a display unit and a user input unit; and a server connected to the user terminal via a wired or wireless network.

[0023] An intelligent autonomous system according to one embodiment of the present invention refers to a system configured to make decisions independently based on information collected from an external environment or an internal state, and to execute physical or logical operations according to the result of the decision. Such an intelligent autonomous system may be implemented as a single device or in a combined form of multiple hardware and software components, and may be utilized in various application fields without being limited to a specific industry or use.

[0024] Intelligent autonomous systems may generally include a family of sensor components for perceiving the external environment. Here, the term "sensor" can be understood as a concept encompassing all forms of sensing means capable of detecting the state of an environment or object, such as physical location, movement speed, acceleration, distance, temperature, illuminance, sound, image, pressure, and biosignals, and may be configured as a single sensor or a combination of multiple sensors. Information collected from sensors may be in analog or digital form and may be acquired in real-time or periodic manner.

[0025] In addition, the intelligent autonomous system may include an information processing area that processes information collected from the sensor and converts it into meaningful data. This information processing area may include various processing steps such as signal processing, data normalization, feature extraction, data combination, and state estimation, and may be implemented through various computational means such as a central processing unit (CPU), a graphics processing unit (GPU), and a neural network processing unit (NPU). The processing methods used may include rule-based computation, statistical computation, machine learning, or artificial intelligence model-based computation, and are not limited to a specific method.

[0026] An intelligent autonomous system may include a judgment domain that performs judgments or decisions based on processed information. The judgment domain may evaluate the current state, predict the future state, select among multiple possible actions, determine priorities, etc., and may operate based on predefined conditional expressions, policy information, learned models, or a combination thereof. Such judgment may be performed in a single step, or it may have a structure in which multiple judgment steps are performed sequentially or in parallel.

[0027] In addition, the intelligent autonomous system may include an execution domain for reflecting the judgment results into actual operations. The execution domain may include all means for performing physical or logical operations, such as motors, actuators, actuators, control signal output units, and communication signal transmission units, and may perform operations such as movement, stopping, rotation, operation, signal output, and data transmission depending on the judgment results. These execution operations may be continuous or may be performed on an event basis.

[0028] The intelligent autonomous system may be implemented in a form where the aforementioned sensors, information processing, decision-making, and execution domains are integrated within a single device, or in a distributed structure interconnected via a network. For example, some functions may be performed internally within the system, while others may be performed in conjunction with external devices or servers, and data may be transmitted and received via wired or wireless communication. In this case, the intelligent autonomous system can operate not only as a standalone system but also in a cooperative or interconnected manner.

[0029] Meanwhile, the form and components of intelligent autonomous systems may vary widely depending on the operating environment, purpose of application, or industrial sector. For example, they can be utilized in various fields such as manufacturing facilities, logistics systems, vehicles, medical equipment, home appliances, information processing devices, and virtual environment systems, and can be applied broadly without being limited to specific embodiments. Therefore, the intelligent autonomous systems mentioned in this specification should not be interpreted as being limited to a specific structure, function, or implementation method, but should be understood as a concept encompassing the entire system capable of autonomous judgment and operation.

[0030] The behavioral unit verification and control according to one embodiment of the present invention refers to a series of procedures and functions that break down an operation to be performed by a system into individually distinguishable behavioral units, rather than focusing on the overall function or result, verify the suitability and safety of each behavior before it is actually executed or during the execution process, and adjust whether or not to execute or the method of execution based on the results. Here, the term "behavioral unit" can be understood as a concept that includes the smallest unit of action in which the system affects the outside or inside, such as physical movement, mechanical operation, control signal output, data processing request, and communication command transmission.

[0031] Behavioral unit verification may include a process of confirming whether an individual behavior defined as above aligns with the system's current state, operational purpose, environmental conditions, and pre-set constraints at the time when such behavior is about to occur. In this case, the scope of verification is not limited merely to the type of behavior or command value, but can comprehensively consider the timing of the behavior, the frequency of its repetition, the sequential order of its occurrence, and the interrelationships between the behaviors. Therefore, behavioral unit verification can be extended to a concept that includes not only judgments on single events but also the evaluation of behavioral patterns over time.

[0032] Behavioral unit verification can be performed according to various criteria, which may include physical feasibility, logical consistency, environmental suitability, system resource status, and the potential for external influence. For example, verification criteria may include whether a specific behavior falls outside the scope of physically implementable capabilities, whether it is permissible under current environmental conditions, and whether there are any inconsistencies in terms of continuity with previous behaviors. These criteria may be defined in advance or may be dynamically adjusted based on information accumulated during system operation.

[0033] Behavioral unit control refers to the process in which the execution of an actual action is adjusted based on the aforementioned verification results. The control method is not limited to allowing the execution of the action as is, but may include forms such as restrictively executing the action by modifying some attributes or parameters, delaying the timing of the action's execution, or suspending or blocking the action itself. Furthermore, control may be performed once for a single action, or it may be performed stepwise by reflecting the accumulated judgment results for multiple actions occurring consecutively.

[0034] Behavioral unit verification and control can be applied equally to systems involving physical operations as well as to logical or information processing-oriented systems. For example, data access requests, requests to perform operations, and requests for communication with external systems can also be defined as behavioral units and subject to verification and control. As such, behavioral unit verification and control is not limited to specific types of systems or operations, but can be understood as a general concept for comprehensively managing all actions that a system is capable of performing.

[0035] Furthermore, behavioral unit verification and control are not limited to structures performed solely within a single system but can be extended to include interactions between distributed components. In this case, behaviors occurring in each component may be verified individually or as a combined result of multiple behaviors. Through such a structure, flexibility to respond to diverse operating environments and requirements can be provided while maintaining the overall stability and predictability of the system.

[0036] Accordingly, the behavioral unit verification and control described in this specification are not limited to specific implementation methods, specific judgment criteria, or specific control techniques, and should be interpreted as encompassing all forms of technical concepts that perform proactive or dynamic verification and control based on individual actions performed by the system.

[0037] An on-device security framework according to one embodiment of the present invention refers to a security structure configured such that all or part of the security functions are performed within the device itself to be protected or within a local computing environment closely linked to the device, without relying on an external server or a centralized system. The basic concept of such an on-device security framework is to enable stable security operation regardless of network connection status or changes in the external environment by allowing functions such as security judgment, verification, control, and logging to be processed independently within the device.

[0038] An on-device security framework can generally be implemented by utilizing computational, storage, and control resources installed within the device. Here, computational resources may include various forms of computational means, such as a central processing unit, auxiliary processing unit, or neural network processing unit, and are capable of a wide range of processing, from simple rule-based processing to complex data computation or model-based decision-making. Storage resources may include memory or storage media for storing security policy information, verification criteria, historical data, etc., and may be configured in a volatile or non-volatile form.

[0039] An on-device security framework can utilize various events, requests, or state changes occurring during the operation of a protected device as input elements for security decisions. These input elements may include control signals, data processing requests, sensor inputs, user inputs, and communication events originating within the device, and can be processed not only as a single element but also in a combined form of multiple elements. This enables security decisions that comprehensively consider the device's current state and operational context, rather than relying on fragmentary information.

[0040] Furthermore, the on-device security framework may have a structure capable of directly reflecting the results of security decisions into the actual operation of the device. For example, control can be implemented by allowing or restricting the execution of specific actions, modifying certain attributes of an action, or delaying or suspending its execution. Such control may be performed immediately or applied in stages depending on whether conditions are met. By linking security decisions and execution control within the same device in this manner, security gaps caused by external delays or communication errors can be minimized.

[0041] The on-device security framework can also internally record and manage information generated during security-related decision-making and control processes. These records may include various information such as the time of event occurrence, the basis for judgment, and control results, and can be utilized as data for device operational history or post-hoc analysis. The recording method can be implemented in various forms, such as sequential storage, periodic updates, and cumulative management, and is not limited to a specific format or structure.

[0042] Meanwhile, the on-device security framework can be applied independently to a single device or extended into a distributed structure where multiple devices interact while each performing their own on-device security functions. In this case, each device can perform its own security decisions while exchanging information with other devices or external systems as needed. Through this structure, the autonomy and independence of individual devices can be secured simultaneously while maintaining a security level at the overall system level.

[0043] Accordingly, the on-device security framework described in this specification is not limited to specific hardware configurations, specific security algorithms, or specific operating methods, and should be interpreted to encompass all forms of security structures and operational concepts configured to perform security functions within a device or local computing environment.

[0044] A user terminal (100) according to one embodiment of the present invention includes, for example, a personal computer, a server computer, a handheld or laptop device, a mobile device (mobile phone, PDA, media player, etc.), a multiprocessor system, a consumer electronic device, a mini computer, a mainframe computer, a distributed computing environment including any of the aforementioned systems or devices, and an edge computing environment in which data is processed at the edge where data is generated rather than at a central server, and the configuration is not limited only to what is described.

[0045] The user terminal (100) may include at least one processor and memory. Here, the processor may include, for example, a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor, an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), and may have multiple cores.

[0046] The memory may be volatile memory (e.g., RAM, etc.), non-volatile memory (e.g., ROM, flash memory, etc.), or a combination thereof. Additionally, the user terminal (100) may include additional storage. The storage includes, but is not limited to, magnetic storage, optical storage, etc. The storage may store computer-readable instructions for implementing one or more embodiments disclosed herein, and may also store other computer-readable instructions for implementing an operating system, an application program, etc. Computer-readable instructions stored in the storage may be loaded into memory to be executed by a processor.

[0047] Additionally, the user terminal (100) may include a user input unit (110) and an output device. The user input unit (110) may include, for example, a keyboard, a mouse, a pen, a voice input device, a touch input device, an infrared camera, a video input device, or any other input device. Additionally, the output device may include, for example, one or more displays, speakers, printers, or any other output devices. Additionally, the computing device may use an input device or an output device provided in another computing device as the user input unit (110) or output device. Additionally, the computing device may include a communication module that enables the computing device to communicate with another device. Here, the communication module may include a modem, a network interface card (NIC), an integrated network interface, a radio frequency transmitter / receiver, an infrared port, a USB connection, or other interfaces for connecting the computing device to another computing device. The communication module may include a wired connection or a wireless connection.

[0048] Each component of the user terminal (100) may be connected by various interconnections such as a bus (e.g., peripheral component interconnection (PCI), USB, firmware (IEEE 1394), optical bus structure, etc.) or may be interconnected by a network. Terms such as “component,” “system,” etc. used in this specification generally refer to computer-related entities that are hardware, a combination of hardware and software, software, or running software.

[0049] A user terminal (100) according to one embodiment of the present invention may include a display unit (120), and the method of implementing the display of the display unit (120) is not limited and may be implemented in various display methods such as liquid crystal, plasma, light-emitting diode, organic light-emitting diode, surface-conduction electron-emitter, carbon nanotube, nanocrystal, etc. In the case of a liquid crystal method, the display unit (120) includes a liquid crystal display panel, a backlight unit that supplies light to the liquid crystal display panel, and a panel driving unit that drives the liquid crystal display panel. Meanwhile, the display unit (120) may be implemented as an OLED panel, which is a self-emissive element, without a backlight unit.

[0050] A server (200) according to one embodiment of the present invention may include a communication unit (210), a storage unit (220), and a control unit (230).

[0051] A communication unit (210) according to one embodiment of the present invention can communicate with a user terminal (100) or other external electronic devices, etc., via wired or wireless communication methods. Therefore, in addition to a connection unit including a connector or terminal for wired connection, it can be implemented in various other communication methods. For example, it can be configured to perform one or more of Wi-Fi, Bluetooth, Zigbee, infrared communication, Radio Control, Ultra-Wide Band (UWM), Wireless USB, and Near Field Communication (NFC). The communication unit (210) may include communication modules such as Bluetooth Low Energy (BLE), Serial Port Profile (SPP), Wi-Fi Direct, infrared communication, Zigbee, and Near Field Communication (NFC). Additionally, the communication unit (210) can be implemented in the form of a device, a software module, a circuit, a chip, etc.

[0052] A communication unit (210) according to one embodiment of the present invention may include various communication modules as described above, and may include an IoT communication module having an IoT network for each carrier. An IoT communication module may refer to any IoT communication network in which a plurality of objects having separate communication units (210) are connected through a network to enable services based on various platforms. By using such an IoT communication module, a smoother communication network can be provided within a set area.

[0053] A storage unit (220) according to one embodiment of the present invention may receive information from a user terminal (100) or from an external search platform through a communication unit (210) and store it, and may receive text information and image information included in a plurality of web pages received by a control unit (230) and store them. The storage unit (220) may store various data according to the processing and control of the control unit (230) described later. The storage unit (220) may be accessed by the control unit (230) to perform reading, recording, modification, deletion, updating, etc. of data. The storage unit (220) may include non-volatile memory such as flash memory, hard disk drive, and SSD (solid-state drive) so as to preserve data regardless of whether system power is provided to the server (200). Additionally, the storage unit (220) may include volatile memory, such as a buffer or RAM, for temporarily loading data processed by the control unit (230).

[0054] A control unit (230) according to one embodiment of the present invention can perform control for the operation of various components of a server (200). The control unit (230) may include a control program (or instruction) that enables such control operation, an inactive memory in which the control program is installed, a volatile memory in which at least a portion of the installed control program is loaded, and at least one processor or CPU (Central Processing Unit) that executes the loaded control program. In addition, such a control program may be stored in an external electronic device other than the server (200).

[0055] The control program may include program(s) implemented in at least one form among BIOS, device driver, operating system, firmware, platform, and application. In one embodiment, the application may be pre-installed or stored in the server (200) at the time of manufacturing the server (200), or may be installed in the server (200) based on the received data by receiving data of the application from an external source when it is used later. The data of the application may be downloaded to the server (200) from an external server, such as an application market, for example, a platform according to the present invention, but is not limited thereto. Meanwhile, the control unit (230) may be implemented in the form of a device, S / W module, circuit, chip, etc., or a combination thereof.

[0056] A control unit according to one embodiment of the present invention is implemented to control the overall operation of an on-device security framework method, and the on-device security framework method includes the step of receiving, through the communication unit, an execution command and an operation request generated from a decision unit of an intelligent autonomous system and transmitted to a physical execution unit.

[0057] The control unit is configured to manage the flow of data entering through the receiving interface of the communication unit, and the receiving step is performed at a logical boundary that ensures the execution command calculated by the decision unit must pass through the communication unit before being transmitted to the execution unit. The communication unit may include at least one of a wired or wireless network interface, an internal bus interface, or an inter-process communication interface, and the control unit detects a receiving event occurring at the said interface to obtain an operation request.

[0058] The above execution command and operation request are not limited to a single command string but may consist of a data set including a command identifier, the time of request creation, identification information of the decision module that created the request, the type of execution unit to be executed, and control parameters required for execution. The control unit sequentially loads the data set received from the communication unit into a buffer and manages each request to be distinguished even when multiple operation requests are received at the same time by assigning a timestamp based on the time of reception. Such timestamps may be generated using an internal system clock or a synchronized reference clock, and may be implemented using a general time function call method used by ordinary technicians.

[0059] The control unit compares the decision module identification information included in the request data with the list of decision modules previously registered in the storage unit to verify whether the received execution command and operation request were actually generated by the decision unit. This comparison can be performed using a simple string comparison method or by checking for a match in the module identifier hash values. For example, the implementation can compare the identifier hash value included in the request data generated by the decision module with the reference hash value stored in the storage unit, classifying it as a valid request only if they match.

[0060] In addition, the control unit may perform a format check based on the length of the requested data, the order of field configuration, and the presence of required parameters to verify whether data is corrupted or abnormally transmitted during the reception of execution commands and operation requests. For example, if an execution parameter field is missing or the length of the received data is shorter than a predefined minimum length, the request may not be treated as a valid execution command but is classified into a separate pending state. This format check consists of inspection logic that can be easily implemented by a person of ordinary technical skill through standard data parsing procedures.

[0061] To verify the feasibility of the receiving phase, the control unit can perform repetitive experiments under conditions identical to the actual operating environment. For example, the decision unit is configured to generate execution commands at regular intervals, and measurements are taken to determine whether each request transmitted through the communication unit is successfully loaded into the control unit's receiving buffer and whether the difference between the reception time and the request generation time remains within an acceptable range. In this case, the reception delay time can be calculated by subtracting the generation time from the reception time, and the stable performance of the receiving phase can be confirmed by calculating the average and variance from multiple experimental results. These experiments are performed repeatedly based on actual data, and the measured results confirm that the receiving phase is not arbitrarily set but is a verified procedure under actual system operating conditions.

[0062] As described above, the step in which the control unit receives execution commands and operation requests through the communication unit is implemented as a prerequisite step for recognizing and managing operation requests prior to physical execution. The composition of the data received at this stage, the reception method, and the verification procedure are not limited to a specific implementation form and can be modified and applied by a person skilled in the art to suit various system environments. Therefore, the receiving step described in this embodiment should be understood as a general technical concept that enables the control unit to acquire execution commands and operation requests before transmitting them to the execution unit.

[0063] An on-device security framework method according to one embodiment of the present invention includes the step of identifying whether the entity that generated the operation request corresponds to an internal software process, an external input signal, or a remote control request, and then repeatedly verifying the validity of the identity and authority at each operation request time by comparing it with the entity-specific identity information and authority information already stored in the storage unit.

[0064] Here, the entity that generated the operation request can be understood as a concept encompassing individual software processes running within the intelligent autonomous system, sources of external input signals flowing in from sensors, switches, buttons, signal lines, etc., or the sender of a remote control request transmitted through a network. The control unit is implemented to distinguish which type of entity generated the request based on the metadata included in the received operation request.

[0065] In the case of an action request generated by an internal software process, the control unit extracts at least one of the process identifier, execution path information, and a unique identification value assigned at the time of process creation included in the request data. Such identifiers can be obtained in ways commonly used by ordinary technicians, such as process IDs provided at the operating system level, task handles, or fixed string values ​​assigned upon application execution. Based on the corresponding identifier, the control unit compares it with internal process-specific identity information already stored in the storage unit to determine whether the process that generated the request is a pre-registered legitimate process.

[0066] When an operation request is generated by an external input signal, the control unit identifies the entity based on the type of interface where the input signal was received, the location where the input signal originated, and the electrical or logical characteristic values ​​of the signal. For example, a signal received from a specific sensor port, a specific input channel, or a specific interrupt line can be identified as the entity corresponding to that port or channel. By comparing this input path information with the identity information for each input entity stored in the storage unit, the control unit determines whether the external input originated from an authorized entity.

[0067] When an operation request is generated by a remote control request, the control unit identifies the subject using at least one of the sender identification information, communication session information, authentication token, or key value included in the request packet received through the communication unit. Such identification information may consist of an IP address, a device unique identifier, a session ID, or an encrypted token string, and the control unit compares this information with the identity information for each remote subject already stored in the storage unit. This comparison process may be implemented in a manner such as checking for simple string matching, comparing hash values, or checking the token validity period.

[0068] After subject identification is completed, the control unit retrieves authorization information corresponding to the subject from the storage unit. The authorization information may include the types of actions the subject can perform, the allowed execution range, the execution frequency, or execution conditions, and this information may be stored in a table, list, or key-value structure. The control unit verifies the validity of the authorization by determining whether the execution target and execution content included in the received action request fall within the range allowed by the authorization information.

[0069] This identity and authorization verification is implemented not to be performed only once during the system's initial setup, but to be executed repeatedly at every operation request. In other words, even if consecutive operation requests occur from the same subject, the control unit independently performs the subject identification and authorization comparison procedures for each request. This ensures that changes in the subject's state, authorization, or the external environment can be reflected immediately.

[0070] To verify the feasibility of the above iterative verification step, the control unit may conduct experiments in an actual operating environment by generating internal processes, external inputs, and remote requests a certain number of times or more, respectively. For example, the same internal process is configured to generate operation requests at regular time intervals, and it is recorded whether identification results and authorization verification results are consistently derived for each request. Additionally, operation requests are artificially generated from unauthorized subjects to verify whether such requests are distinguished during the identification or authorization comparison phases. By accumulating and verifying the actual collected request data and the corresponding verification results, this experiment confirms that the subject identification and iterative verification procedures are reliably performed based on actual data.

[0071] As described above, the step of identifying the entity that generated the operation request and repeatedly verifying the validity of the identity and authority at each operation request point by comparing it with the identity and authority information already stored in the storage unit is not limited to a specific identification method or a specific authority model, and can be implemented in various ways by a person skilled in the art to suit the system environment. Therefore, the verification step described in this embodiment should be understood as a general technical concept for continuously verifying the reliability of operation requests according to the entity type.

[0072] An on-device security framework method according to an embodiment of the present invention includes the step of generating first context data including the system location, movement speed, time information, presence of surrounding workers, environmental data collected from multiple sensors, and the temporal change amount of each sensor data at the time when the operation request occurs, and identifying second context data representing the state of the current execution environment by sorting the first context data by component and combining them according to mutual correlation.

[0073] Here, the first context data is a basic data set intended to represent the execution environment at the time an operation request is generated from various angles, and can be understood as data in which information of different natures is collected and organized based on the same point in time. When the control unit receives an operation request, it sets the point in time corresponding to the request as the reference point, and constructs the first context data by acquiring information corresponding to each component within a certain time interval before and after that reference point.

[0074] System location information refers to the spatial coordinates where the intelligent autonomous system is located at the time of an operation request. In indoor environments, it can be expressed as relative coordinates relative to a fixed reference point, and in outdoor environments, as satellite-based or map coordinate values. The control unit receives location values ​​from a location sensor, map matching information, or an internal location estimation module, and stores them as numeric coordinate data. Movement speed information can be derived using the change in the location information over time or values ​​measured by a separate speed sensor; for example, it can be calculated by dividing the difference in distance between location coordinates measured at regular time intervals by time.

[0075] Time information is data representing the absolute or relative time at which an operation request occurs, and is generated through an internal system clock or a synchronized reference clock. The control unit uses this time information as a reference field for the first context data to enable data collected from different sources to be aligned on the same time axis. This allows position, velocity, and sensor data to be combined into a context at the same point in time, even if they have different collection cycles.

[0076] The presence of a nearby worker is information indicating whether a person exists around the system, and can be determined by at least one of camera-based recognition results, proximity sensor detection results, or whether a worker tag signal is received. The control unit converts this input into a binary value or a step value and includes it in the first context data. For example, if a worker is detected, it is represented by a specific value, and if not detected, by a different value, thereby enabling comparison during a subsequent combination process.

[0077] Environmental data collected from multiple sensors may include data representing the state of the surrounding environment of the system, such as temperature, illuminance, vibration, noise, and acceleration. The control unit stores the measurement values ​​periodically received from each sensor for a certain period before and after a reference point in time, calculates a representative value within that period, and includes it in the first context data. This representative value may be set as one of an average value, a maximum value, or a measurement value at a specific point in time, and a person skilled in the art may select it based on the characteristics of the sensor.

[0078] The temporal variation of each sensor data is calculated using the difference between continuously collected measurements from the same sensor. For example, the difference between the sensor's previous and current measurements can be calculated and defined as the variation, where the magnitude of the variation can be expressed as a simple difference value or a value obtained by applying the sqrt() operation to the sum of squares. This variation can be used as an indicator to determine whether the environment is in a static state or a rapidly changing state.

[0079] The control unit generates first context data by configuring the collected location, movement speed, time information, worker presence, environmental data, and temporal change amounts into independent fields, and then performs sorting by data type. Sorting can be performed according to predefined criteria, such as the order of time information, spatial information, and environmental information, thereby arranging different types of data into a consistent structure. This process can be implemented in the form of a simple array or a table, allowing a person of ordinary skill to easily perform it.

[0080] Subsequently, the control unit identifies the second context data by combining data based on the interrelationships between each component of the aligned first context data. An interrelationship refers to a relationship that verifies whether multiple context elements simultaneously satisfy specific conditions, such as when a movement speed exceeds a certain value and the presence of a worker is detected at the same time. The control unit evaluates these relationships using a condition comparison method to represent the state of the current execution environment as a single data structure. In this case, the second context data is generated not as a simple set of individual sensor values, but as identification data representing the comprehensive state of the execution environment.

[0081] To verify the feasibility of the above-mentioned first and second context data generation processes, the control unit may conduct experiments in which various positions, speeds, and sensor conditions are artificially changed in an actual operating environment. For example, the system is operated in both a stationary and a moving state, and operation requests are repeatedly generated while varying the presence of surrounding workers. Subsequently, the first context data generated at each point in time and the second context data identified therefrom are recorded to determine whether they are consistently distinguishable. By accumulating and managing these records together with the actual collected sensor data, it can be confirmed that the context data generation and combination procedures are stably performed based on actual data.

[0082] The step of generating first context data as described above, sorting it by component, and combining it according to mutual relationships to identify second context data is not limited to a specific sensor type or a specific combination algorithm, and can be implemented by a person skilled in the art by modifying it in various ways depending on the system environment and the purpose of application. Therefore, the context data generation and identification steps described in this embodiment should be understood as general technical concepts for quantitatively expressing the state of the execution environment at the time of the operation request.

[0083] An on-device security framework method according to one embodiment of the present invention calculates a risk level corresponding to an operation request using a neural network processing unit (NPU) with the identity and authorization verification result and the second context data as input values, wherein the risk level includes a value calculated stepwise by comparing it with allowable range data included in a predefined security policy and physical safety regulation.

[0084] Here, the neural network processing unit is a computational resource having a parallel processing structure distinct from a general central processing unit, and is implemented to perform operations on multiple input values ​​simultaneously. The control unit converts the identity and authorization verification results and the second context data into an input format that the NPU can process and transmits them as input values. In this case, the identity and authorization verification results may be expressed as binary values ​​or stepwise numeric values ​​to distinguish between cases where the subject is verified as normal and cases where it is in a restricted or unverified state, and the second context data consists of multiple numeric elements representing the state of the current execution environment.

[0085] The control unit normalizes the range of input values ​​in advance so that each element input to the NPU can substantially contribute to the calculation of risk. For example, the results of identity and authorization verification represent an allowed state as 0 and a restricted state as 1, and values ​​related to location, speed, or environmental status included in the second context data are linearly transformed within the range of minimum and maximum values ​​of actual measurement data collected during system operation. This normalization process can be implemented through simple numerical calculations and can be easily performed by a person of ordinary skill using a standard data preprocessing method.

[0086] The NPU calculates a risk level corresponding to an action request using the aforementioned input values. This risk level is not a single operation result, but rather a value that reflects the combined state of identity and authorization verification results and secondary context data, and is calculated to vary depending on the combination of input values. For example, even for the same action request, the risk level may be calculated to be higher if the subject's authorization status is restricted, or if risk factors such as the presence of nearby workers or increased movement speed are included in the secondary context data. This operation can be implemented by repeatedly performing multiplication and summation operations between weights and input values ​​within the NPU.

[0087] The calculated risk level is compared with allowable range data included in predefined security policies and physical safety regulations. Allowable range data consists of numerical ranges set according to the type of operation request, system operating environment, or safety standards, and is stored in advance in the storage unit. The control unit compares the risk level output from the NPU with the corresponding allowable range to determine which level the risk level belongs to. This leveling calculation can be classified, for example, into a low level if the risk level is within the first range, a medium level if it exceeds the second range, and a high level if it exceeds the third range; this classification is performed through a simple conditional comparison operation.

[0088] The stepwise risk calculation process can be verified based on data collected from actual operating environments. For example, various identity states and different second context data conditions are artificially generated to repeatedly trigger operation requests, and the risk values ​​calculated under each condition are recorded. By verifying whether the recorded risk values ​​are classified into consistent levels based on comparisons with acceptable range data, it can be confirmed that the risk calculation and grading process operates stably based on actual data. This verification process can be implemented by cumulatively managing the input data and output results collected during the experiment.

[0089] As described above, the step of calculating the risk level using an NPU with the results of identity and authorization verification and the second context data as inputs, and performing stepwise computations by comparing this with allowable range data included in predefined security policies and physical safety regulations, is not limited to a specific neural network structure or a specific risk calculation method. A person skilled in the art can implement this by modifying the configuration of input elements, normalization methods, and criteria for separating steps according to the system environment and the purpose of application. Therefore, the risk calculation step described in this embodiment should be understood as a general technical concept for quantitatively evaluating the risk level of an operation request.

[0090] An on-device security framework method according to one embodiment of the present invention includes the step of controlling the transmission of an operation request to an execution unit when the calculated risk level is less than or equal to a first threshold value, controlling the performance of at least one restriction control among reducing the operation speed, limiting the output, or changing the path when the risk level exceeds the first threshold value and is less than or equal to a second threshold value, and controlling the transmission of the operation to the execution unit or suspending the operation until an external approval input is received when the risk level exceeds the second threshold value.

[0091] Here, the control unit receives the risk level calculated in the previous step as a single numerical value and sequentially compares that numerical value with the first and second reference values ​​already stored in the storage unit. The first and second reference values ​​serve as boundary values ​​to distinguish between different risk levels; they are set according to security policies and physical safety regulations and can be managed as multiple sets of values ​​depending on the type of operation request or the system operation mode. Based on the identification information included in the operation request, the control unit queries the set of reference values ​​applicable to the request and uses it for the comparison operation.

[0092] If the calculated risk level is determined to be below the first threshold value, the control unit determines that the operation request falls within an acceptable range under the current execution environment and identity / authorization status, and transmits the operation request to the execution unit without modification. At this time, transmission may be performed by at least one of a control signal, a command packet, or an internal interface call, and can be implemented using a general control signal transmission structure used by a person of ordinary skill. The control unit records the time of transmission and the identifier of the transmitted request together to enable tracking of the execution result thereafter.

[0093] If the calculated risk level is determined to exceed a first threshold value and be less than or equal to a second threshold value, the control unit does not immediately block the operation request itself but switches to a restricted control state. Restricted control includes at least one of reducing the operation speed, limiting the output, or changing the route, and is implemented by modifying the control parameters included in the operation request. For example, in the case of movement operation, the target speed value is reduced by a certain ratio, or in the case of output control, the maximum output value is replaced with a limited upper limit. In the case of route change, route information can be modified to select an alternative route by excluding sections of the existing route that contain risk factors. Such restricted control can be implemented using simple numerical substitution or conditional branching logic, allowing a person skilled in the art to easily perform it.

[0094] If the calculated risk level exceeds the second threshold value, the control unit determines that executing the corresponding operation request under current conditions poses a significant physical or security risk and blocks the transmission of the operation to the execution unit or switches it to a hold state. In the case of blocking, the control signal itself is not generated to be transmitted to the execution unit; in the case of holding, the operation request is temporarily stored in an internal queue or storage unit, and execution is delayed until an external approval input is received. The external approval input may be received from a management device, an operator terminal, or a higher-level control system; upon receipt of such input, the control unit re-evaluates the hold operation request or transmits it to the execution unit.

[0095] The control unit may conduct experiments to artificially generate different risk conditions in order to verify each of the above control results in an actual operating environment. For example, for the same operation request, the risk input value is varied stepwise, and the form of the control signal transmitted to the execution unit, whether restriction control is applied, and whether it is blocked or suspended are recorded under each condition of being below a first threshold value, exceeding the first threshold value and below a second threshold value, and exceeding the second threshold value. These records are accumulated and managed together with the actual execution log, thereby confirming that control branching operations based on risk are performed consistently.

[0096] As described above, the step of controlling operation requests to execute them as is, apply restriction controls, or block or suspend them according to the risk value is not limited to a specific threshold value setting method or a specific restriction control method. A person skilled in the art can implement this by varying the number and range of threshold values ​​and the specific form of restriction controls depending on the characteristics of the system, applicable safety regulations, or the operating environment. Therefore, the risk-based control step described in this embodiment should be understood as a general technical concept for differentially controlling the execution intensity of operation requests according to the calculated risk level.

[0097] An on-device security framework method according to one embodiment of the present invention includes a step of controlling the generation of log data in an immutable form, wherein the time of receipt of the operation request, the result of identity and authorization verification, the composition of the second context data, the result of risk calculation, the history of control execution, and the final execution status are sequentially stored in the storage unit.

[0098] Here, the control unit determines the time at which an operation request is received through the communication unit as the reference time and records that time as the first field of the log entry. The time of reception can be acquired with a resolution of at least milliseconds using an internal system clock or a synchronized reference clock, and a request identifier is assigned to distinguish cases where multiple requests occur at the same time.

[0099] The results of identity and authorization verification are recorded exactly as the judgment values ​​derived from the preceding verification step. Specifically, judgment results such as subject identification, authorization approval status, and restriction status are expressed as binary or stepwise values ​​and included in log entries. Since the values ​​recorded here use the output of the verification logic as is, no additional interpretation or conversion is required, and a standard technician can easily implement this by simply mapping the verification result variables directly to the log fields.

[0100] The composition of the second context data serves as a core element representing the state of the execution environment at the time of the operation request, and the control unit includes the value or identification result of each element constituting the second context data in the log entry. For example, the values ​​for each field constituting the second context data, such as location category, movement status, worker presence, and environmental status classification results, are recorded as is. In this process, it is not necessary to store the entire raw sensor data; the size of the log can be managed by recording only the result values ​​used to identify the second context data.

[0101] The risk calculation result consists of the final risk value output through NPU computation and level information indicating which level the value belongs to, and is recorded in the log. The control unit immediately reflects the value calculated immediately after the risk calculation in the log entry, ensuring that the initial calculation result is preserved even if the value is subsequently changed or recalculated. At this time, the risk value and level information can be recorded as simple numerical values ​​or code values, allowing an ordinary technician to implement it without separate complex calculations.

[0102] The control execution history is information indicating the type of control action selected by the control unit based on the risk comparison results; it is recorded by clearly distinguishing whether it corresponds to execution delivery, restriction control, blocking, or suspension. In the case of restriction control, the control item actually applied—such as operation speed reduction, output limitation, or path change—is also recorded. This allows for retrospective verification of which control was applied, even for the same operation request.

[0103] The final execution status is recorded as a value indicating whether the operation request was actually transmitted to the execution unit and executed, executed in a restricted form, or not executed at all, based on the aggregated results of the control execution. The control unit determines the final execution status based on whether it was transmitted to the execution unit or whether external approval input was received, and saves the corresponding status as the last field of the log entry.

[0104] The above log data is generated in an immutable form. To this end, the control unit stores each log item sequentially by concatenating them, and can provide continuity by including the identification value or hash value of the previous log item in the next log item. For example, by generating a new hash value by combining the hash value of the previous log item with the content of the current log item and storing them together, it is possible to implement a system where continuity with subsequent logs is broken if an intermediate log item is modified. This method can be implemented using only general hash operations and string concatenation, making it easy for a person skilled in the art to execute.

[0105] Log data is stored sequentially in the storage unit according to the order of creation. The control unit controls the recording of log entries to the next storage location in the storage unit whenever a new entry is created, and does not allow the modification or deletion of existing log entries. The storage unit can be implemented as non-volatile memory, a file system, or a storage medium capable of sequential writing, and the recording method can also be sufficiently implemented using a simple append method.

[0106] To verify the feasibility of the above-mentioned log generation and storage steps, the control unit repeatedly generates various operation requests in an actual operating environment and checks whether the log items generated for each request are recorded sequentially without omission, covering information from the time of reception to the final execution status. Additionally, by conducting experiments in which some log items are artificially modified or deleted, the immutability of the logs can be verified by checking whether an anomaly is detected during the continuity verification process. This verification can be performed repeatedly based on actual collected log data.

[0107] As described above, the step of generating and sequentially storing immutable log data for the entire process from the time of receiving an operation request to the final execution status is not limited to a specific log format or storage technology. A person skilled in the art can implement the configuration of log items, the method of granting continuity, and the storage structure by modifying them in various ways depending on the system environment and the characteristics of storage resources. Therefore, the log generation and storage steps described in this embodiment should be understood as general technical concepts for reliably recording key judgments and control processes occurring in the on-device security framework method.

[0108] A control unit according to one embodiment of the present invention generates first change amount data by calculating a time-unit change amount for each type of sensor for a plurality of sensor data constituting the second context data, and generates second change amount judgment data by cross-comparing the first change amount data between sensors to determine whether it deviates from a change range that can be physically established simultaneously.

[0109] Here, the multiple sensor data constituting the second context data refers to data collected from sensors measuring different physical quantities to represent the execution environment at the time of the operation request, and may include, for example, position sensors, speed sensors, acceleration sensors, vibration sensors, illuminance sensors, distance sensors, or environmental state sensors. Each sensor data is stored in chronological order according to a fixed period or the time of event occurrence, and the control unit performs the calculation of the amount of change using pairs of data continuously collected from the same sensor.

[0110] The change in time unit for each type of sensor is calculated based on the difference between the value at the previous measurement point and the value at the current measurement point for the same sensor. For example, in the case of a position sensor, the distance traveled can be calculated using the difference in position coordinates between the two points in time and divided by the measurement time interval to define the change. In the case of acceleration sensors or vibration sensors, the difference between the previous and current measurements can be calculated as an absolute value, or in the case of multi-axis data, the difference between each axis value can be squared, summed, and then processed using the sqrt() operation to produce a single change value. Since these operations can be implemented through simple arithmetic, a person of ordinary skill can easily perform them without specialized knowledge.

[0111] The control unit generates first change amount data by bundling the change amounts for each sensor type calculated as above with the sensor identifiers. The first change amount data may be composed of a data structure in which the change amount values ​​for each sensor are sorted in chronological order, and each item includes the sensor type, the change amount value, and the calculation reference point. Through this, change amounts calculated from different sensors are organized so that they can be compared based on the same point in time.

[0112] Subsequently, the control unit cross-compares the multiple sensor change amounts included in the first change amount data. Cross-comparison is a process of determining whether change amounts calculated from different sensors fall within a range where they can physically occur simultaneously. It refers to identifying combinations that are difficult to establish simultaneously in terms of physical causality, such as cases where the position change amount increases rapidly while there is almost no movement speed, or where the surrounding environment sensor value does not change at all despite a rapid increase in the vibration change amount. To make this determination, the control unit uses comparison criteria that reflect the physical relationship between the sensors.

[0113] The aforementioned range of physically simultaneous changes can be set based on actual sensor data collected during system operation. For example, the range of simultaneous occurrence between each sensor change amount is identified from sensor data collected over a certain period during normal operation, and that range is stored in the storage unit as a reference value. Subsequently, a judgment is made by comparing whether the first change amount data calculated in real time falls outside that range. At this time, the comparison can be performed by a simple threshold comparison or a determination of whether it falls within the range, making implementation easy.

[0114] If the control unit determines, based on the cross-comparison results between sensors, that one or more combinations of sensor change amounts fall outside the range where they can physically be simultaneously established, it generates the corresponding determination result as second change amount determination data. The second change amount determination data may consist of stepwise determination values, such as whether it is within the normal range, the abnormal range, or a suspected state, and may also include identification information regarding which sensor combination caused the deviation from the range. Through this, the change amount determination result can be linked with other control or verification logic in subsequent steps.

[0115] To verify the feasibility of the above-mentioned change amount calculation and cross-comparison process, the control unit may conduct experiments in which a system equipped with actual sensors is operated in both normal and abnormal states. For example, data may be collected by artificially changing only some sensor inputs while maintaining the system in a stationary state, or by rapidly changing specific sensor values ​​while the system is in motion. It is then verified whether the first change amount data and the second change amount judgment data generated at this time are created by distinguishing between the normal and abnormal states. These experiments are repeated based on actual sensor data, and through the results, it can be confirmed that the change amount judgment logic operates by reflecting actual physical characteristics rather than being an arbitrary setting.

[0116] The step of generating second change amount judgment data by calculating the change amount per time unit for each sensor type as described above, and determining whether it exceeds the range of physically simultaneous change that can be established through cross-comparison between sensors, is not limited to a specific sensor configuration or a specific comparison criterion. A person skilled in the art can implement this by adjusting the change amount calculation method and cross-comparison criteria in various ways depending on the physical characteristics of the system, sensor placement, and operating environment. Therefore, the change amount-based judgment step described in this embodiment should be understood as a general technical concept for verifying the temporal consistency of multiple sensor data.

[0117] A control unit according to one embodiment of the present invention controls the calculation of risk by classifying the sensor data as unreliable data when, in the second change amount judgment data, specific sensor data is identified as exhibiting mutually contradictory change patterns within the same time interval compared with other sensor data, thereby progressively reducing the weight or excluding it from the calculation target when calculating the risk.

[0118] Here, the second change amount judgment data refers to data that includes the judgment result regarding whether the change amounts calculated from multiple sensors are within a range where they can physically exist simultaneously; it is data in the form of mutual comparison results organized by time interval, along with the change amount values ​​for each sensor.

[0119] A mutually contradictory change pattern within the same time interval refers to a case where a combination of changes that is difficult to occur simultaneously due to general physical laws or system operation characteristics is observed between sensors measuring related physical quantities. For example, this may apply to cases where a sudden change in position is produced in position sensor data simultaneously with speed sensor data judged to maintain a moving speed close to zero, or where a large change is produced in acceleration sensor data simultaneously with vibration sensor data judged to be in a state with almost no shock or vibration. Whether such a contradiction exists is determined by aligning the change amounts between sensors within the same time interval and comparing them.

[0120] The control unit checks whether specific sensor data exhibits a contradictory change pattern repeatedly or more than a certain number of times when compared with other sensor data, based on the combination of change amounts for each sensor included in the second change amount judgment data. In this case, repetition can be determined based on whether the same sensor shows an inconsistent change pattern compared to other sensors over multiple consecutive time intervals. For example, if a contradictory pattern is observed in a certain proportion or more of the total number of comparisons, the corresponding sensor can be identified as an unreliable candidate. Since this judgment criterion can be implemented using a simple counting method or a ratio comparison method, it can be easily performed by a person skilled in the art.

[0121] Classification as unreliable data is a step intended to ensure that sensor data in which contradictory patterns have been identified as described above is not treated at the same reliability level during the risk calculation process. The control unit can control the weight applied to the sensor data during risk calculation to be reduced stepwise. For example, for sensor data where a weight of 1.0 was applied in a normal state, the weight can be adjusted by gradually lowering it, such as to 0.7 if a contradictory pattern is detected one or more times, and to 0.4 if it is detected multiple times. This stepwise reduction can be implemented using a simple table reference or a conditional branching method.

[0122] In addition, the control unit can control the system to completely exclude the relevant sensor data from the risk calculation target if the degree of inconsistency pattern exceeds a certain standard. For example, if the same sensor data repeatedly exhibits a change pattern that is physically incompatible with other sensors over multiple consecutive time intervals, the sensor data is determined to have lost reliability due to a temporary error or external interference and can be excluded from the input value for risk calculation. In this case, since the risk calculation is performed using only other sensor data, it is possible to prevent a single sensor error from having an excessive influence on the overall judgment result.

[0123] The aforementioned criteria for weight reduction or exclusion from calculation can be established based on sensor data collected in an actual system operating environment. For example, sensor data collected over a long period under normal operating conditions is compared and analyzed with data collected under conditions where sensor errors or external interference are intentionally introduced to identify the correlation between the frequency of contradiction patterns and system malfunctions. Based on these analysis results, it can be determined at what level of contradiction pattern occurrence it is reasonable to adjust the weight or exclude the data.

[0124] To this end, the control unit can inject artificial noise into a specific sensor in an experimental environment or temporarily destabilize the sensor connection to cause a change amount different from the actual value, and then verify the comparison results with other sensor data. The second change amount judgment data and risk calculation results generated during this process are recorded in the storage unit, and by verifying the relationship with normal operation, it can be confirmed that the logic for classifying unreliable data and adjusting weights is a method verified by actual data.

[0125] As described above, the configuration for determining whether specific sensor data exhibits mutually contradictory change patterns within the same time interval when compared with other sensor data, classifying the said sensor data as unreliable data, and controlling the calculation by progressively reducing its weight or excluding it from the calculation target when calculating risk is not limited to a specific type of sensor or a specific system structure. A person skilled in the art can implement this by setting various contradiction judgment criteria, weight adjustment steps, and exclusion conditions according to the sensor configuration, system characteristics, and operating environment, and this embodiment specifically explains such general implementation concepts.

[0126] A control unit according to one embodiment of the present invention controls the risk level by recalculating the risk level using only adjusted sensor data, thereby preventing an operation request generated by sensor deception, data contamination, or environmental disturbance from being transmitted to a physical execution stage.

[0127] Here, adjusted sensor data refers to a data set composed of values ​​corrected according to the confidence level, incorporating sensor data that was previously classified as unreliable and had its weight reduced or excluded from calculation, thereby selecting only reliable sensor data.

[0128] When recalculating the risk, the control unit uses only sensor data that is not classified as unreliable from the entire sensor data originally used for the risk calculation, or uses the correction value of the weighted sensor data as the input. In this case, the risk recalculation process follows the same procedure as the existing risk calculation, but since only the input data set is changed, a person of ordinary skill can easily implement it without making separate, complex structural changes. For example, it can be implemented by controlling the input to the operation block that calculates the risk by excluding specific sensor items from the array of sensor data or by inputting a value obtained by multiplying those items by a pre-calculated correction coefficient.

[0129] Operation requests resulting from sensor deception may occur when a specific sensor outputs a value different from the actual environment due to the injection of external signals, forged inputs, or the intentional induction of errors. Data contamination includes cases where sensor data is deformed into abnormal values ​​due to communication errors, memory corruption, or temporary malfunctions of the sensor itself. Environmental disturbances include cases where the sensor output fails to reflect the actual state due to external environmental factors such as strong electromagnetic interference, vibration, or sudden temperature changes. Considering that if the risk level is calculated using the sensor data generated in such situations as is, there is a possibility that an operation request that is actually dangerous may be judged as normal or low-risk and transmitted to the physical execution stage, the control unit is configured to use only adjusted sensor data.

[0130] The results of the risk recalculation can be adjusted upward or downward compared to the risk prior to adjustment. For example, if a specific sensor is deceptive in a way that lowers the risk by generating an excessive safety signal, excluding that sensor data will result in a relative increase in the risk. Conversely, if a specific sensor is contaminated in a way that raises the risk by generating an excessive danger signal, excluding that sensor data or reducing its weight will result in the risk being recalculated to a value closer to the actual situation. Since the risk recalculated in this way reflects the reliability of the sensor data, it is more suitable for determining whether to transmit it to the physical execution stage.

[0131] The control unit determines whether to transmit the operation request to the execution unit based on the recalculated risk level. If the recalculated risk level exceeds a preset allowable range, the operation request is blocked or kept in a pending state, thereby effectively preventing operation requests generated by sensor deception, data contamination, or environmental disturbance from being transmitted to the physical execution stage. Since this control can be implemented using conditional branching or state transitions based on the risk calculation results, it can be easily performed by a person skilled in the art.

[0132] The risk recalculation procedure based on the aforementioned adjusted sensor data can be verified by data collected in an actual operating environment. For example, the risk can be calculated using sensor data collected in a normal environment as well as sensor data collected by artificially injecting noise into a specific sensor or generating external interference. By comparing the risk change before and after adjustment with the actual system operation results, it can be confirmed that the recalculated risk is suitable for determining physical execution safety. In this process, the sensor data, the risk calculation results, and whether actual execution occurred are recorded together, thereby proving that the risk recalculation using only adjusted sensor data is valid based on actual data.

[0133] The configuration in which the control unit recalculates the risk using only the sensor data adjusted as described above, and controls whether to transmit the physical execution of an operation request based on the result, is not limited to a specific type of sensor, a specific risk calculation algorithm, or a specific physical execution unit. A person skilled in the art can apply various modifications to the sensor data adjustment method and the risk recalculation method depending on system characteristics and application fields, and this embodiment has been described in sufficient detail to enable such general implementation.

[0134] A control unit according to one embodiment of the present invention generates first control data by decomposing the control parameters included in the operation request by component, and compares the first control data with pre-set physical guardrail data item by item, wherein the physical guardrail data is configured to include at least one limit value among a maximum rotation angle, a maximum acceleration, an allowable range of movement, and a continuous operation duration.

[0135] Here, the control parameters included in the operation request refer to a set of information expressed as numerical or state values ​​regarding the direction, magnitude, speed, and temporal duration conditions of the operation to be performed by the physical execution unit, and the control unit decomposes these parameters into rotation-related parameters, movement-related parameters, speed and acceleration-related parameters, time-related parameters, etc., to form the first control data.

[0136] When generating the first control data, the control unit extracts each parameter field included in the operation request message or control frame as is, or, in the case of composite parameters, converts them into a single physical quantity unit and stores them. For example, if a rotation command is provided as a combination of angle and rotation time, the control unit can separate this into a rotation angle value and a rotation speed value and include them in the first control data. Since this decomposition process can be implemented through simple data parsing and unit conversion, it can be easily performed by a person skilled in the art.

[0137] Physical guardrail data refers to a data set that defines the range within which the system or device can operate safely in a structural and physical manner, in the form of limit values. The maximum rotation angle indicates the maximum angle at which a physical execution part can rotate relative to a specific axis, and this may be determined by joint structures, wiring structures, or interference conditions regarding the device's external shape. The maximum acceleration indicates the acceleration limit that a drive unit, reducer, or coupling structure can withstand, and the allowable travel range refers to the range of coordinates movable within the workspace or device installation space. The continuous operation duration refers to the maximum time during which continuous operation is possible, taking into account heat generation, wear, or power consumption.

[0138] The control unit compares each item of the first control data with the corresponding item of the physical guardrail data on a one-to-one basis. For example, it determines whether the rotation angle value included in the first control data exceeds the maximum rotation angle limit value, and whether the acceleration value exceeds the maximum acceleration limit value. In the case of a movement command, it can compare and determine whether the movement end coordinates calculated based on the current position fall within the allowable movement range, and in the case of a continuous operation duration, it can compare whether the requested operation time exceeds the set maximum duration time. Such comparisons can be implemented through simple numerical comparisons or determinations of range inclusion.

[0139] The limit values ​​for physical guardrail data are not set arbitrarily but can be derived based on actual device testing or operational data. For example, through repetitive rotation tests, the point at which structural interference or abnormal vibration occurs above a certain angle can be identified, and the maximum rotation angle can be set to the value prior to that point. In the case of acceleration, operational tests are performed by gradually increasing acceleration, and the limit value can be set based on the point where a sudden surge in motor current, temperature rise, or control instability begins to occur. The duration of continuous operation can also be calculated based on the point at which heat generation and performance degradation occur through long-term continuous operation tests. The control parameters and operational status data collected during these testing processes are stored and can be utilized as the basis for setting physical guardrail data.

[0140] The control unit determines whether the control parameters included in the operation request satisfy all the limit values ​​of the physical guardrail data by utilizing the comparison results for each item. If even some items exceed the limit values, the operation request may be determined to be physically unsafe and can be used as input information for restriction, correction, or blocking control in subsequent stages. This configuration is not limited to a specific device structure or control method and is generalized so that the same concept can be applied to various physical execution units.

[0141] A control unit according to an embodiment of the present invention controls the entire operation request to be invalidated when any of the first control data is identified as exceeding the limit value of the physical guardrail data, or to generate reconstructed second control data by modifying only the control parameter corresponding to the excess item, and then re-enter it into the risk calculation step.

[0142] Here, the control unit maintains the result of comparison with physical guardrail data for each item of the first control data, and if any single item is determined to exceed a threshold value, this is used as a trigger condition. This determination does not simply refer to the result of a single threshold comparison, but can be implemented as a logical judgment confirming whether at least one of multiple items has exceeded the allowable range.

[0143] Invalidating an entire operation request refers to a situation where the control unit determines that there is a high probability of causing structural damage, compromising safety, or system malfunction if the request is transmitted to the physical execution unit. In this case, the control unit does not forward the operation request to subsequent processing steps, but classifies the request as discarded and excludes it from the execution flow. This process can be implemented by assigning an identifier to the operation request and marking that identifier as invalid, and can be easily implemented in software by a person skilled in the art.

[0144] Meanwhile, the control unit may generate second control data by modifying only the control parameters corresponding to the excess items, rather than uniformly invalidating operation requests in all excess situations. In this case, the control unit identifies items in the first control data that exceed a limit value and replaces the value of the corresponding item with a limit value included in the physical guardrail data or a value less than or equal to the limit value. For example, if the rotation angle value exceeds the maximum rotation angle, the control unit may replace the corresponding rotation angle with the maximum rotation angle value or reduce it to a value obtained by multiplying the maximum rotation angle by a certain ratio and reflect this in the second control data. In the case of acceleration or travel distance, only the excess items can be adjusted in the same manner.

[0145] This process of modifying control parameters is not performed arbitrarily, but is based on limit values ​​established using data collected during actual device testing or operation. For example, if a section where vibration or control instability occurs above a certain rotation angle is identified through repetitive operational testing, the value of the section immediately preceding that point can be set as a reference value, and the control logic can be configured to automatically adjust to that reference value when the angle is exceeded. Since both the control parameter values ​​and the operational results are recorded during this testing process, it can be confirmed that the modification criteria have been verified by actual data.

[0146] The control unit generates second control data containing modified control parameters and re-enters it as the input value for the risk calculation step. That is, the risk calculation is performed again based on the second control data, which has been adjusted to satisfy the physical guardrail, rather than the original first control data. This prevents operation requests exceeding physical limit values ​​from being directly reflected in the risk calculation and execution judgment.

[0147] According to this structure, the control unit performs stepwise control by normalizing or correcting operation requests based on physical limits and then re-evaluating them, rather than a binary approach of simply allowing or blocking operation requests. This is not dependent on specific control algorithms or device structures and can be implemented as a general control concept applicable to various intelligent autonomous systems and physical execution units.

[0148] A control unit according to an embodiment of the present invention controls at least one of an abnormal or excessive control command generated by a decision unit and an external input to be blocked before it is transmitted to an execution unit. Here, a control command generated by a decision unit refers to a control signal calculated according to a computational result within an intelligent autonomous system, and a control command generated by an external input refers to a signal generated by user input, a remote control signal, or a control request received from an external system. The control unit receives these control commands at an intermediate stage before transmitting them to the execution unit and determines whether the control command is normal.

[0149] An abnormal control command refers to a control command in which the range of change in control parameters is significantly large compared to the normal operating range performed repeatedly under the same conditions, or which exhibits a pattern of rapid increase or decrease within a short time interval. For example, if the rotation angle or output value increases or decreases excessively compared to the previous average value under the same operating conditions, the control unit may classify this as an abnormal control command. This determination can be made not based solely on the value at a single point in time, but through a comparison with the history of control commands accumulated over a recent period, and can be easily implemented by a person of ordinary skill using a log-based comparison or a sliding window method.

[0150] An excessive control command refers to a case where the control command itself has a normal format, but its size or duration exceeds the physical characteristics of the execution unit or the usual operating range. For example, if the continuous operation time exceeds a generally acceptable range, or if the output rises rapidly, potentially causing an overload on the execution unit, the control unit identifies this as an excessive control command. In this case, the control unit can determine whether the numerical value included in the control command exceeds a preset range by comparing it with the preset range, and this comparison criterion may be based on a value set by analyzing the control command data collected during actual device operation and the execution results.

[0151] If the control unit determines that a control command is abnormal or excessive, it blocks the command from being transmitted to the execution unit. This blocking does not merely mean blocking the communication path, but includes logically branching the control command's processing flow to prevent it from entering the stage of transmission to the execution unit. For example, a status flag can be assigned to the control command to distinguish between an executable state and a blocked state, and the control flow can be configured so that commands classified as blocked are not transmitted to the execution unit.

[0152] Such blocking criteria are not set arbitrarily but can be derived based on control commands collected during actual system operation and the resulting execution data. For example, by repeatedly collecting data on the magnitude, rate of change, and duration of control commands during normal and abnormal operation periods, and comparing and analyzing this data, boundary values ​​can be established to distinguish between abnormal or excessive control commands. This process can be verified by repeatedly applying the same control command in an experimental environment and observing the response of the execution unit; the results indicate that the established criteria reflect the actual operational characteristics.

[0153] Accordingly, the control unit is implemented to determine whether a control command generated from a decision unit or external input is normal or excessive before it is transmitted to the execution unit, and to block control commands identified as abnormal or excessive, thereby preventing in advance a situation in which the execution unit performs unintended operations or suffers physical damage.

[0154] A control unit according to an embodiment of the present invention generates first behavioral history data by accumulating execution history, risk calculation results, and actual control execution results over a past period for operation requests repeatedly received from the same entity. Here, the same entity refers to any one of an internal software process, an external input device, a remote control terminal, or a user account distinguished by identity information identified in the preceding step, and the control unit classifies operation requests based on entity identification information included in each operation request.

[0155] The past period refers to a period corresponding to a recent specific time interval or a recent specific number of times among the operation request history accumulated and stored in the storage unit by the control unit. For example, at least one of operation requests received during the last 10 minutes, operation requests performed 100 times during the last 100 minutes, or execution records during the last day can be selected and set as the reference period, and such a reference can be set by considering the frequency of operation requests actually collected and the execution cycle in the system operating environment. The control unit sequentially checks whether to execute operation requests included in the reference period and correlates the risk value calculated in response to each operation request with the control result actually performed.

[0156] Execution history refers to information including whether an operation request was transmitted to the execution unit and actually performed, and whether restrictive control was applied or blocked. Risk calculation result refers to the risk value or risk range information calculated in the preceding step for the corresponding operation request, and actual control execution result refers to information indicating whether control was carried out in the form of execution, restricted execution, or blocking depending on the risk level. The control unit combines this information by operation request to form a single history item and accumulates history items repeatedly generated for the same entity in chronological order.

[0157] The first action history data may consist of a data set in which these history items are arranged along a time axis, and each item may be configured to include at least the time of occurrence of the action request, the result of risk calculation, and the result of control execution. The control unit may manage the size of the history data by updating the first action history data corresponding to the subject whenever a new action request is received, and by determining whether to remove or preserve the oldest history item. This method can be easily implemented by a person skilled in the art using a circular buffer, queue, or list structure.

[0158] Since this first behavioral history data accumulates and reflects operation requests and their results that occur repeatedly during actual system operation, it naturally includes the operation patterns of specific entities, trends in risk level changes, and the distribution of control results. For example, if the same entity repeatedly generates operation requests classified as high-risk, records in which the risk level is calculated to be above a certain level may continuously accumulate in that entity's first behavioral history data. Because these accumulated results are formed based on data collected in the actual execution environment, it can be recognized that they are data derived from actual operation history rather than arbitrarily set values.

[0159] Accordingly, the control unit generates first behavioral history data by accumulating execution history, risk calculation results, and control execution results for a certain period for operation requests repeatedly received from the same entity, and is implemented to stably construct history data reflecting the operational characteristics of each entity through this.

[0160] A control unit according to an embodiment of the present invention identifies behavioral patterns for each subject based on the frequency of operation requests, average risk values, and restriction or blocking occurrence rates from the first behavioral history data. If it is determined that the behavioral pattern deviates from the pattern range typically allowed within the authority scope of the subject, the control unit controls the subsequent operation requests generated by the subject by lowering the reference threshold value when calculating the risk or by expanding the risk range to which restriction control is applied, thereby thereby suppressing the possibility of authority misuse or insider attacks by a subject that has undergone normal authentication in stages.

[0161] A control unit according to an embodiment of the present invention identifies behavioral patterns for each subject based on the frequency of operation requests, the average risk value, and the rate of restriction or blocking occurrences from the first behavioral history data. Here, the frequency of operation requests refers to a value obtained by converting the number of operation requests received from the same subject over a certain period into a time unit, and can be calculated, for example, as the number of requests per minute or the number of requests per hour. The average risk value can be defined as a value obtained by summing the risk values ​​calculated for the operation requests of the subject during the said period and dividing by the number of requests, and the rate of restriction or blocking occurrences refers to the ratio of requests classified as restriction control or blocking among the total operation requests. These values ​​can be calculated without separate external data by utilizing the execution history and risk calculation results already accumulated and stored in the first behavioral history data.

[0162] The control unit organizes the calculated operation request frequency, average risk value, and restriction or blocking occurrence rate into a set of behavioral characteristics specific to each entity, and compares this with the typical behavioral patterns corresponding to the authority range of the entity. Here, the range of patterns typically allowed within the authority range refers to the distribution of values ​​derived from behavioral history collected during the past normal operation periods of entities with the same authority level or the same entity. For example, the ranges of the average operation request frequency, average risk value, and restriction occurrence rate under normal operation conditions can be calculated in advance, and then it can be determined whether the current behavioral characteristics of a specific entity fall outside this range. This determination can be performed by considering multiple indicators simultaneously rather than relying on a single indicator, and a person skilled in the art can implement this using critical interval comparison or multiple condition comparison methods.

[0163] Cases where the aforementioned behavioral pattern is determined to deviate from the range of patterns typically permitted within the authority scope of the relevant entity include, for example, instances where the frequency of operation requests consistently exceeds the upper limit of the normal range, where the average risk value rises significantly compared to past normal intervals, or where the rate of restriction or blocking increases above a certain level and is repeatedly observed. These judgment criteria can be repeatedly verified based on the first behavioral history data collected during actual system operation and can be established based on the point where the normal state and the abnormal state are clearly distinguished for a specific entity.

[0164] For entities determined to exhibit abnormal behavioral patterns as described above, the control unit either lowers the threshold value applied during risk calculation for subsequent operation requests or expands the risk range to which restriction control is applied. Lowering the threshold value means changing the criteria so that restriction or blocking control is applied at a lower risk value than before, even under the same input conditions. Furthermore, expanding the risk range to which restriction control is applied means broadening the scope to include certain risk levels previously classified as "execution" within the restriction control range. Such adjustments may be applied exclusively to the relevant entity and configured so as not to affect the control criteria of other entities.

[0165] Such adjustment of criteria is not based on arbitrary judgment, but can be derived by repeatedly verifying the trends in execution history, risk calculation results, and control execution results accumulated in the actual first behavioral history data. For example, by observing changes in the frequency of restrictions or blocking and system stability while gradually lowering the threshold value, and reflecting the results back into the first behavioral history data, the control intensity corresponding to the behavioral characteristics of the relevant entity can be naturally adjusted.

[0166] Therefore, even if a subject has undergone normal authentication, if its behavioral pattern deviates from the usual pattern within the scope of authority, the control unit can control the possibility of authority misuse or insider attack by gradually adjusting the risk calculation criteria and the scope of restriction control application in stages, thereby allowing the possibility of authority misuse or insider attack to be gradually suppressed according to the accumulated behavioral history.

[0167] A control unit according to one embodiment of the present invention sorts log generation target data, including the time of receipt of the operation request, identity and authorization verification results, the composition of the second context data, the risk calculation result, and the control execution result, in chronological order, and then generates first log feature data by calculating waveform-based feature values ​​from the log generation target data at each log generation time.

[0168] A control unit according to an embodiment of the present invention sorts log generation target data, including the time of receipt of the operation request, the result of identity and authorization verification, the composition of the second context data, the result of risk calculation, and the result of control execution, in chronological order. Here, log generation target data refers to result values ​​that have already been calculated and confirmed throughout the entire process of each operation request being processed, and refers to a data set that can be secured in a storage unit without separate external input or additional computation. The control unit sorts the data by assigning a common time standard to each data item and arranging them in the order of occurrence, thereby preserving the temporal flow between consecutive operation requests.

[0169] The control unit constructs a waveform for the log generation target data aligned as described above, which includes the trend of data changes before and after each log generation point. In this context, a waveform refers to a set of values ​​that change continuously over time along a time axis, such as changes in risk calculation results, state transitions of control execution results, or repetitive patterns of identity and authorization verification results. For example, the pattern of risk calculation results increasing or decreasing over time, or patterns of repetitive alternation between restriction control and execution control, can be represented as a waveform.

[0170] The control unit calculates waveform-based feature values ​​from the waveform. A waveform-based feature value is a value that quantifies the morphological characteristics of a waveform and may include, for example, at least one of the magnitude of the difference between consecutive risk values, the frequency of change within a certain time interval, or the degree of dispersion of the value. Such feature values ​​can be derived through simple arithmetic operations and can be implemented, for example, by accumulating the difference in values ​​between adjacent time intervals or by applying a sqrt() operation to the sum of squares of the change amount within a certain interval to calculate the fluctuation intensity. Such operations can be easily implemented by a person skilled in the art using only log data arrays and basic mathematical operations.

[0171] The control unit generates first log feature data by configuring waveform-based feature values ​​calculated at each log generation point into a single data item. The first log feature data may consist of a set of waveform-based feature values ​​generated in chronological order, and each item summarizes the characteristics of the operation request processing at the corresponding log generation point. For example, waveform-based feature values ​​may be calculated as relatively large at points where the risk value fluctuates rapidly, and as relatively small values ​​during periods where stable control is sustained.

[0172] Since such first log feature data is directly calculated from log generation target data repeatedly collected during actual system operation, it can be seen that it is data based on actual operation history rather than arbitrarily assumed values. For example, if log data is collected by comparing cases where normal operation is repeated in the same environment with cases where abnormal control situations occur, the difference in the distribution of waveform-based feature values ​​corresponding to the two cases can be clearly identified, thereby demonstrating that the first log feature data is formed by actual data.

[0173] Accordingly, the control unit is implemented to stably construct log characteristic data that reflects the temporal change characteristics of the entire operation request processing process by sorting the data to be logged in chronological order and generating first log characteristic data by calculating waveform-based characteristic values ​​at each log generation point.

[0174] A control unit according to one embodiment of the present invention calculates the waveform-based feature value by mapping each item of the data to be logged to be generated into a continuous signal on a time axis, and including an amplitude value corresponding to the amount of data change between adjacent logs and a period value corresponding to the interval between log generation points, and by combining the amplitude value and the period value to form a unique waveform pattern for each log point.

[0175] A control unit according to one embodiment of the present invention maps each item of the data subject to log generation into a continuous signal on a time axis when calculating the waveform-based feature value. Here, mapping into a continuous signal on a time axis means arranging values ​​corresponding to each item of the data subject to log generation based on the log generation time point, thereby expressing them in the form of a signal in which values ​​change continuously over time. For example, changes in the risk calculation result, the state value of the control execution result, or the identity and authorization verification result can be arranged as values ​​corresponding to each log generation time point to form a single continuous signal.

[0176] The control unit calculates an amplitude value corresponding to the amount of data change between adjacent logs for the log generation target data mapped to the continuous signal. An amplitude value is a value indicating how much a data value has changed between two consecutive log generation points in time; for example, it can be expressed as a single value by calculating the absolute difference between values ​​at adjacent points in time or by summing the amount of change of multiple items. This calculation of the amount of change can be implemented through simple arithmetic operations, and a person skilled in the art can easily perform it by calculating the difference between adjacent items for the data arrangement for each log point in time.

[0177] In addition, the control unit calculates a period value corresponding to the interval between log generation times. The period value is a numerical value representing the time interval during which consecutive logs are generated, and can be calculated, for example, using the difference in timestamps between two log generation times. In this case, if the log generation cycle is constant, the same period value may be calculated repeatedly; if the log generation cycle is irregular, different period values ​​may be calculated for each interval. Since this period value accurately reflects the time interval during which logs are generated in the actual system operating environment, it can be calculated based on actual collected time information without any separate assumptions.

[0178] The control unit calculates waveform-based feature values ​​by combining the amplitude and period values ​​calculated as above to form a unique waveform pattern for each logarithmic point in time. The combination of amplitude and period values ​​means configuring the waveform so that even if the amplitude value is the same, different waveform characteristics are expressed when the period value is different, or so that even if the period value is the same, the intensity of waveform fluctuation is expressed differently depending on the magnitude of the amplitude value. For example, cases where a large amplitude value is repeated within a short time interval and cases where a small amplitude value occurs within a long time interval can be distinguished as different waveform patterns.

[0179] Since such waveform-based feature values ​​are based on amplitude and period values ​​directly calculated from the data subject to log generation, they do not rely on external variables unrelated to stress or physical quantities. In other words, the waveform-based feature values ​​are formed by reflecting the temporal variation characteristics of the log data itself and are configured to have a unique waveform pattern for each log point in time. During actual system operation, when normal operation is repeated, the combination of amplitude and period values ​​remains stable within a certain range; however, in the event of abnormal control situations or requests for aggressive operation, it may be observed that at least one of the amplitude or period values ​​appears significantly different from the existing pattern.

[0180] Accordingly, the control unit maps the data to be logged as a continuous signal on the time axis, calculates an amplitude value corresponding to the amount of data change between adjacent logs and a period value corresponding to the interval between log generation times, and combines these to form a unique waveform pattern for each log time point, thereby implementing the waveform-based feature value to be specifically derived based on the temporal characteristics of the actual log data.

[0181] A control unit according to one embodiment of the present invention continuously combines the first log feature data with the first log feature data generated at a previous log point to form a cumulative waveform sequence, and controls the identification of whether the log data is changed based on whether the continuity of the cumulative waveform sequence is damaged or whether a discontinuous phase change occurs in a specific section.

[0182] A control unit according to one embodiment of the present invention forms a cumulative waveform sequence by continuously combining the first log feature data with the first log feature data generated at a previous log point in chronological order. Here, "continuously combining" means sequentially concatenating the first log feature data generated at each log point without interruption to form a single set of time-continuous waveform data. This can also be implemented by simply arranging them according to the chronological relationship of the log generation times. A person skilled in the art can easily implement this by storing the feature values ​​generated for each log point in the form of a list, array, or stream and accumulating them in chronological order.

[0183] Since the accumulated waveform sequence formed by the control unit has a structure in which waveform-based feature values ​​at individual log time points are continuously connected along the time axis, the phase, amplitude, and period changes of the waveform maintain constant continuity when logs are generated and accumulated normally. In other words, no temporal or numerical discontinuity occurs between the end of the waveform at a previous log time point and the beginning of the waveform at a subsequent log time point, and the amount of change also takes the form of a gradual continuation.

[0184] The control unit determines whether continuity is compromised in such a cumulative waveform sequence. Compromise of continuity means that the connection relationship between adjacent first log feature data on the cumulative waveform sequence is abruptly severed beyond the normal range of change, or that the flow of the waveform appears in a form where it cannot be physically continued. For example, even though the time interval between consecutive log points is within the normal range, if the amplitude value changes abruptly to a level completely unrelated to the previous pattern, or if waveforms in intermediate sections are removed or traces of duplication appear on the cumulative waveform sequence, it may be determined that continuity is compromised.

[0185] In addition, the control unit identifies whether the log data has been modified based on whether a discontinuous phase change occurs in a specific section of the accumulated waveform sequence. A discontinuous phase change refers to a state in which the direction of propagation, periodic flow, or reference point of the waveform in the accumulated waveform sequence moves abruptly, resulting in a non-natural connection with the previous waveform. This is a phenomenon that is unlikely to occur when log data is generated and accumulated normally, and it is mainly observed when data at a specific log point in time is arbitrarily modified, deleted, or inserted.

[0186] To determine whether such discontinuous phase changes exist, the control unit calculates the phase difference between adjacent waveform segments on the accumulated waveform sequence and verifies whether the difference falls outside the range observed in experimentally verified normal log accumulation environments. For example, in a normal log accumulation experimental environment, the amount of waveform phase change between consecutive logs is repeatedly measured to secure the distribution range; then, by calculating the amount of phase change in the same manner from the accumulated waveform sequence generated during actual operation and comparing them, it is possible to determine whether the phase change in a specific segment is discontinuous. Since this process can be repeatedly verified based on actual log data, it can be sufficiently reproduced by a skilled technician in the same log collection environment.

[0187] In this way, the control unit is configured to accumulate first log feature data in chronological order to form a waveform sequence, and to identify whether the log data has been modified based on whether the continuity of the accumulated waveform sequence is damaged or whether a discontinuous phase change occurs in a specific section. Since this identification of log modification is performed based on the temporal and waveform continuity of the actual generated and accumulated log data itself, without relying on external reference values ​​or arbitrary judgment factors, it is possible to determine specifically and reproducibly whether the log data has been maintained normally or if there has been a modification in the interim.

[0188] A control unit according to an embodiment of the present invention stores waveform linkage information generated based on the accumulated waveform sequence and original log data in correspondence, wherein the waveform linkage information is recorded in a storage unit so as to be reused as a reference value when calculating waveform-based feature values ​​of logs generated thereafter, thereby configuring the system so that waveform continuity of logs generated thereafter is not maintained if any of the past logs are modified or deleted, and accordingly, the log data is controlled so that tampering is identified not by hash verification at the single log unit level, but by the physical irreversibility of the waveform continuous structure accumulated over time.

[0189] A control unit according to one embodiment of the present invention records waveform linkage information generated based on the accumulated waveform sequence and log source data in a storage unit so as to correspond with each other. Here, waveform linkage information refers to information indicating how waveform-based feature values ​​for each log point in time constituting the accumulated waveform sequence are connected and formed in chronological order, and is data configured to identify what continuous relationship the waveform at a specific log point in time has with the waveforms at the preceding and subsequent log points in time. Log source data is data that includes actual log generation target data as is, such as the time of receipt of an operation request, identity and authorization verification results, context data, risk calculation results, and control execution results, and is stored in a 1:1 or time-point correspondence relationship with the waveform linkage information.

[0190] The control unit does not merely store waveform linkage information as a historical record, but records it in the storage unit so that it can be reused as a reference value when calculating waveform-based feature values ​​for subsequently generated logs. That is, whenever a new log is generated, the control unit is configured to calculate the waveform-based feature value corresponding to the newly generated log based on the premise that it naturally connects with the existing accumulated waveform sequence, using the waveform state at the most recent point in time among the existing waveform linkage information accumulated in the storage unit as a reference. This process can be performed by referring to the relationship between the terminal characteristics of the previous waveform and the amplitude and period values ​​calculated from the new log, and a person skilled in the art can easily implement this by recalling the stored waveform linkage information as a reference value and reflecting it in the calculation.

[0191] Under this structure, if any past log is modified or deleted, the correspondence between the waveform linkage information recorded in the storage and the actual original log data is broken. In particular, since the waveform linkage information is used as a reference value when calculating waveform-based feature values ​​for subsequent logs, if a part of the past log is changed, the continuity of the accumulated waveform sequence, which should be formed based on that point in time, cannot be maintained. As a result, the waveforms of subsequently generated logs cannot be naturally connected to the existing waveform linkage information, and discontinuous changes occur in amplitude, period, or phase flow.

[0192] The control unit controls the system to identify whether log data has been tampered with by detecting the breakdown of this waveform continuity. In this case, the verification of log data is not performed by comparing hash values ​​generated for individual log units, but is based on the physical characteristics of the waveform continuity structure itself, which accumulates over time. Since the accumulated waveform sequence has a structure in which subsequent waveforms are formed sequentially based on the waveform state of the previous log point, changes to past logs affect all subsequent waveform formation processes, making it practically impossible to restore them to their original state.

[0193] This physical irreversibility can be verified through repeated experiments using actual log data. For example, if specific past logs are artificially modified or deleted after logs have been sequentially generated normally to accumulate waveform linkage information, a phenomenon is reproduced where the waveform-based feature values ​​of logs generated at subsequent points in time do not match the existing waveform linkage information. Conversely, it can be confirmed that waveform continuity is maintained even when the same generation procedure is repeated when the logs remain unchanged. Through these experimental verifications, it is validated that the waveform continuity structure possesses the characteristic of responding sensitively to changes in logs.

[0194] Consequently, the control unit stores waveform linkage information and original log data in correspondence and is configured to reuse the waveform linkage information as a reference value when generating subsequent logs. This ensures that the falsification of log data is identified by the physical irreversibility of the continuous waveform structure formed cumulatively over time, rather than relying on static verification at the single-log unit level. Since this configuration is designed so that changes to the log data affect the entire cumulative structure, it is sufficiently implementable by a person of ordinary skill in a real system environment while ensuring high reliability in detecting falsification.

[0195] A control unit according to one embodiment of the present invention generates third state data by collecting system state change data, including the internal system computation load, power consumption change amount, processing delay time, and memory access frequency at the time the log data is generated, at each time the log data is generated.

[0196] Here, the point in time when log data is generated refers to the point in time immediately before or simultaneously with the recording of the result of a control decision made regarding an operation request, and the control unit is configured to acquire the internal operation and resource usage status of the system based on that point in time.

[0197] The system's internal computational load is a value representing the computational throughput being performed by the control unit at the time of log generation, and can be derived from one or more indicators, such as, for example, the number of computation requests processed during a certain time interval, processor utilization, or the number of processing targets loaded in the job queue. A person of ordinary skill in the field can quantify and acquire the current computational load through a standard monitoring interface provided by the operating system or runtime environment, and the control unit records this value in correspondence with the time of log generation so that the computational load is linked with the log data.

[0198] The change in power consumption is data representing the change in the system's power usage around the time of log generation; the control unit can calculate this by determining the difference in power consumption between the current time and the previous reference time. In this case, power consumption can be obtained through the system's internal power management module, power sensors, or consumption indicators that can be estimated via software, and the change can be implemented by calculating the difference in consumption between two consecutive time points through simple arithmetic operations. This method enables the quantitative reflection of the impact of specific log generation on the system's power status.

[0199] Processing latency refers to the time elapsed from the moment an operation request is received until the control decision regarding that request is completed and log data is generated. The control unit can calculate processing latency by assigning timestamps at the time of request reception and log generation, respectively, and calculating the time difference between them. This value can be compared with the average latency processed by the system in a normal state and can be utilized as an indicator that specifically reflects the system response characteristics at the time of log generation.

[0200] Memory access frequency refers to the number of times system memory is accessed or the frequency of access events during a certain period before and after the time of log generation. The control unit can calculate memory access frequency by aggregating indicators such as the number of memory read and write requests, the number of cache misses, or page access events, and this is included in the third state data as state information representing the density of internal system data processing activities.

[0201] The control unit collects the system's internal computational load, changes in power consumption, processing delay time, and memory access frequency as independent items, and organizes these items to correspond to the same log generation time to form a single third state data. In this process, since each item is stored mapped to the same time standard or the same log identifier, the system state at the time specific log data was generated can be accurately reproduced or analyzed retrospectively.

[0202] The process of generating such third-state data is performed repeatedly in the actual system environment, and consistency between data is maintained by applying the same collection procedure at every log generation point. For example, by comparing the third-state data generated when the system is in a normal state with the third-state data generated when abnormal load or delay occurs for the same type of operation request, it is possible to verify how system state fluctuations change along with the log data. Through this iterative collection and comparison, it can be empirically confirmed that the third-state data actually reflects the system state at the time of log generation, enabling ordinary technicians to easily perform system implementation and verification based on this.

[0203] A control unit according to one embodiment of the present invention arranges the third state data in chronological order and calculates an irreversible state transition index based on the direction of state change and the magnitude of change between adjacent log generation points, wherein the irreversible state transition index is configured so that it cannot be reproduced identically even if the same log data is used, if the physical state of the system at the time of log generation is different.

[0204] A control unit according to an embodiment of the present invention arranges the third state data in chronological order based on the log generation time to form a continuous state column. Here, since the third state data is composite state data including internal system computational load, changes in power consumption, processing delay time, and memory access frequency collected at each log generation time, the control unit is configured to align it on a time axis while maintaining the same item order and unit system. This provides a basis for directly comparing how the physical state of the system has changed between two consecutive log generation times.

[0205] The control unit correlates third-state data between adjacent log generation points to calculate the direction and magnitude of change for each state item. The direction of change can be identified by whether it increases or decreases compared to the previous point in time, and the magnitude of change is calculated through the numerical difference between the two points in time. For example, in the case of computational load, the difference between the computational load value at the previous log generation point and the computational load value at the subsequent log generation point can be calculated to determine the direction of increase or decrease and the absolute value of that difference as the magnitude of change. Changes in power consumption, processing latency, and memory access frequency are also derived for the direction and magnitude of state changes, respectively, by calculating the difference between adjacent points in the same manner.

[0206] The control unit generates an irreversible state transition index by combining the direction and magnitude of change for each item calculated as described above. This irreversible state transition index is not a simple change for a single state item, but rather an indicator representing how the physical operational state of the entire system has continuously shifted between log generation points. In other words, since change patterns observed simultaneously across multiple state items are reflected as a single transition, it becomes difficult to reproduce the same transition index even if only the state value at a specific point in time is restored.

[0207] In this case, the irreversible state transition index is configured to yield different values ​​if the physical state of the system at the time of log generation differs, even if the same log data is generated. For example, even if the same operation request and the same control result are recorded in a log, the pattern of state change between adjacent time points will differ depending on whether the log was generated when the system computational load was high and power consumption had increased rapidly, or when it was generated when the computational load was low and power consumption was stable. Consequently, the irreversible state transition index calculated in the two cases will not match exactly, and the same index cannot be generated again unless the past state sequence is reproduced exactly.

[0208] The control unit performs the process of calculating these irreversible state transition indicators using the same procedure whenever log generation is repeated, ensuring that state transition characteristics accumulated according to consecutive log generation times are naturally formed. Even if the same log generation request is repeated in an actual system environment, internal system computational load, changes in power consumption, processing delays, and memory access patterns differ even slightly; therefore, the change patterns of the third state data arranged in chronological order are not reproduced identically every time. Through such iterative observations, it is confirmed that the irreversible state transition indicators are calculated based on the actual physical state changes of the system, and a person skilled in the art can achieve the same effect by implementing this procedure exactly as is.

[0209] A control unit according to one embodiment of the present invention records the irreversible state transition indicator in a storage unit by correlating it with log data, and subsequently controls the identification of whether the creation time and continuity of the log are maintained based on whether the state transition indicator recalculated from the current system state at the time of log verification and the previously stored state transition indicator, and whether both indicate the same transition path. Accordingly, even if past log data is duplicated, moved, or modified, the same state transition indicator is not reconstructed unless the physical state transition characteristics of the system at the time of log creation are reproduced at the current time, thereby controlling the structural exclusion of subsequent forgery or re-insertion of log data.

[0210] A control unit according to one embodiment of the present invention records the irreversible state transition indicator in a storage unit by correlating it one-to-one with each log data. At this time, the storage unit stores the original log data itself along with the irreversible state transition indicator calculated from the third state data at the time the log was generated, linked by the same identifier, and the storage order of the log data is maintained to reflect the chronological order of the actual generation time. Accordingly, specific log data is managed not as a simple text or event record, but as a complex record unit that includes the system's physical state transition characteristics at the time of generation.

[0211] Subsequently, when the time for log verification arrives, the control unit collects third-state data again based on the current system state and recalculates the state transition index by calculating the direction and magnitude of change between adjacent states according to the same procedure as at the time of log generation. Since this recalculation process is performed based on the actual internal system computational load, power consumption changes, processing latency, and memory access frequency at the time of verification, regardless of the content of the stored log data, the same state transition index will not be derived if the physical state of the system at the time of past log generation is not identical to the current state.

[0212] The control unit compares the recalculated state transition indicator with the state transition indicator already stored in the storage unit to determine whether the two represent the same transition path. Here, the same transition path refers to a case where the direction of state change and the pattern of change magnitude correspond identically in chronological order between consecutive log generation points; this determination is based on whether the overall state transition flow is consistently maintained, rather than the matching of a single item value. Based on the result of this comparison, the control unit can identify whether the corresponding log data has been preserved in a state of continuity from the actual generation point to the present, or whether external interventions such as duplication, movement, or modification have occurred in the interim.

[0213] For example, even if log data generated in the past is simply duplicated on a file-by-file basis and moved to a different location, or if some log entries are modified, the physical state transition characteristics of the system at the time of log generation are not reproduced identically in the current system state. Consequently, the recalculated state transition indicator does not match the stored state transition indicator, and based on this, the control unit determines that the continuity of the log at the time of generation has been compromised. Conversely, if the log has been maintained continuously without any modification or re-insertion since generation, the state transition indicator recalculated at the time of log verification will show the same transition path as the stored state transition indicator.

[0214] Due to this structure, the log verification of the present embodiment is performed based on the system's physical state transition characteristics at the time of log generation, unlike simple hash value comparison or checksum verification. Therefore, even if past log data is duplicated, moved, or modified, it becomes structurally impossible to reconstruct the same state transition indicator unless the system's physical state transition characteristics at the time of log generation are reproduced identically at the present time. As a result, subsequent forgery, re-insertion, or artificial manipulation of log data is naturally excluded during the verification stage, and a person skilled in the art can achieve the same verification effect by implementing the procedure described in this specification exactly as it is. Explanation of the symbols

[0215] 100: User terminal 110: User Input Section 120: Display section 200: Server 210: Communications Department 220: Storage section 230: Control unit

Claims

Claim 1 An on-device security framework method for verifying and controlling behavioral units of an intelligent autonomous system comprises: a user terminal including a display unit and a user input unit; and a server connected to the user terminal via a wired or wireless network, wherein the server comprises a communication unit; a storage unit; and a control unit; wherein the control unit is implemented to control the overall operation of the on-device security framework method, and the on-device security framework method comprises the steps of: receiving execution commands and operation requests through the communication unit before they are generated from the decision unit of an intelligent autonomous system and transmitted to the physical execution unit; identifying whether the entity generating the operation request corresponds to any one of an internal software process, an external input signal, or a remote control request, and then repeatedly verifying the validity of the identity and authority at each operation request time by comparing it with the entity-specific identity information and authority information stored in the storage unit; generating first context data including the system location, movement speed, time information, presence of surrounding workers, environmental data collected from multiple sensors, and the temporal change amount of each sensor data at the time the operation request occurs, and identifying second context data representing the state of the execution environment at the time the operation request occurs by sorting the first context data by component and combining them according to mutual correlations; and calculating a risk level corresponding to the operation request using a neural network processing unit (NPU) with the identity and authority verification results and the second context data as input values, wherein the risk level is A step of performing calculations to include values ​​calculated stepwise by comparing with allowable range data included in predefined security policies and physical safety regulations;A method for an on-device security framework for behavioral unit verification and control of an intelligent autonomous system, characterized by comprising: a step of controlling to transmit the operation request to an execution unit when the calculated risk level is less than or equal to a first threshold value, controlling to perform at least one restriction control among operation speed reduction, output restriction, or path change when the risk level exceeds the first threshold value and is less than or equal to a second threshold value, and controlling to block the transmission of the operation to the execution unit or to suspend the operation until an external approval input is received when the risk level exceeds the second threshold value; and a step of controlling to generate log data in an immutable form, including the time of receipt of the operation request, identity and authorization verification results, the composition of the second context data, the risk level calculation results, control execution history, and final execution status, and to sequentially store it in the storage unit. Claim 2 In claim 1, the control unit generates first change amount data by calculating a time-unit change amount for each type of sensor for a plurality of sensor data constituting the second context data, generates second change amount judgment data by cross-comparing the first change amount data between sensors to determine whether it deviates from a change range that can be physically established simultaneously, and if a specific sensor data is identified in the second change amount judgment data as exhibiting a mutually contradictory change pattern within the same time interval when compared with other sensor data, controls the sensor data to classify it as untrusted data and gradually reduce its weight or exclude it from the calculation target when calculating the risk, and controls the risk to prevent an action request generated by sensor deception, data contamination, or environmental disturbance from being transmitted to the physical execution stage by using only sensor data not classified as untrusted or by using a correction value of the sensor data with adjusted weight. Claim 3 In paragraph 2, the control unit generates first control data by decomposing the control parameters included in the operation request by component, and compares the first control data with pre-set physical guardrail data item by item, wherein the physical guardrail data is configured to include at least one limit value among a maximum rotation angle, maximum acceleration, allowable movement range, and continuous operation duration, and if any of the first control data is identified as exceeding the limit value of the physical guardrail data, the entire operation request is invalidated, or only the control parameters corresponding to the excess item are modified to generate reconstructed second control data and then re-entered into the risk calculation step, thereby controlling to block at least one of the abnormal or excessive control commands generated by the decision unit and external input before they are transmitted to the execution unit. This characterizes an on-device security framework method for behavioral unit verification and control of an intelligent autonomous system. Claim 4 In paragraph 3, the control unit generates first behavioral history data by accumulating execution history, risk calculation results, and actual control execution results over a past period for operation requests repeatedly received from the same entity, and identifies behavioral patterns by entity based on the frequency of operation requests, average risk value, and restriction or blocking occurrence rate from the first behavioral history data; and if it is determined that the behavioral pattern deviates from the pattern range typically allowed within the authority range of the entity, controls subsequent operation requests generated by the entity to lower the reference threshold value when calculating the risk or to expand the risk range to which restriction control is applied, thereby controlling the possibility of authority misuse or insider attack by an entity that has undergone normal authentication to be suppressed in stages. Claim 5 In paragraph 4, the control unit sorts the log generation target data, including the reception time of the operation request, the identity and authorization verification result, the composition of the second context data, the risk calculation result, and the control execution result, in chronological order, and then generates first log feature data by calculating waveform-based feature values ​​from the log generation target data at each log generation time point, wherein the waveform-based feature values ​​are configured to include an amplitude value corresponding to the amount of data change between adjacent logs and a period value corresponding to the interval between log generation times after mapping each item of the log generation target data to a continuous signal on a time axis, and are calculated to form a unique waveform pattern for each log time point by combining the amplitude value and the period value, and the control unit continuously combines the first log feature data with the first log feature data generated at the previous log time point to form a cumulative waveform sequence, and controls to identify whether the log data has changed based on whether the continuity of the cumulative waveform sequence is damaged or whether a discontinuous phase change occurs in a specific section, and stores the waveform linkage information generated based on the cumulative waveform sequence and the original log data accordingly, wherein the waveform linkage information is the waveform of the log generated thereafter An on-device security framework method for behavioral unit verification and control of an intelligent autonomous system, characterized by being configured such that waveform continuity of subsequently generated logs is not maintained if any of the past logs are modified or deleted by recording them in a storage unit so as to be reused as reference values ​​when calculating base feature values, and accordingly, the log data is controlled so that tampering is identified not by hash verification at the single log unit level, but by the physical irreversibility of the waveform continuous structure formed cumulatively over time. Claim 6 In claim 5, the control unit collects system state change data, including the system internal computational load, power consumption change amount, processing delay time, and memory access frequency at the time of log generation, at each time the log data is generated, to generate third state data; arranges the third state data in chronological order, and calculates an irreversible state transition index based on the direction and magnitude of state change between adjacent log generation times, wherein the irreversible state transition index is configured so that it cannot be reproduced identically even if the log data is the same if the system physical state at the time of log generation is different; the control unit records the irreversible state transition index in a storage unit by correlating it with the log data, and subsequently controls the identification of whether the log generation time and continuity are maintained by comparing the state transition index recalculated from the system state at the time of log verification with the previously stored state transition index at the time of log verification, based on whether both indicate the same transition path; accordingly, even if past log data is duplicated, moved, or modified, the same state transition index is not reconstructed unless the system physical state transition characteristics at the time of log generation are reproduced at the time of log verification, thereby thereby post-log data On-device security framework method for behavioral unit verification and control of an intelligent autonomous system, characterized by controlling to structurally exclude forgery or re-insertion.

Citation Information

Patent Citations

  • Humanoid robot somatosensory control system and method based on deep learning

    CN120395905B

  • Agricultural robot path navigation method based on vision-language model

    CN120721088A

  • Unmanned ship automatic control system and method based on plug-in architecture

    CN121348896A

  • Decision architecture for autonomous systems

    US20190318265A1