Man-machine bidirectional self-adaptive jacking device early warning system based on multi-mode pre-judgment

Abstract

The invention discloses a man-machine bidirectional self-adaptive jacking device early warning system based on multi-mode pre-judgment. The man-machine bidirectional self-adaptive jacking device early warning system comprises the following steps that 1, a pre-judgment type sensing layer synchronously collects personnel pre-judgment state trend data representing personnel risks and jacking device fault evolution trend data reflecting equipment faults; step 2, identifying the fault risk of the jacking device by a personnel pre-judgment state sensing module based on the fault evolution trend data of the jacking device; based on the personnel pre-judgment state trend data, the personnel fatigue state and the risk area mistakenly entering are judged; step 3, the man-machine collaborative decision-making layer divides risk levels based on the jacking scenarized dynamic fusion result of the two types of trend data, and generates a two-way protection strategy of personnel risk-device adaptation and device risk-personnel guidance; and step 4, the hierarchical linkage execution layer executes the special gradient early warning and protection action of the jacking operation according to a bidirectional protection strategy and generates and guides a personalized safe evacuation path.

Description

A human-machine bidirectional adaptive early warning system for lifting devices based on multimodal prediction Technical Field

[0001] This invention relates to the field of safety protection technology for jacking devices, and in particular to a human-machine two-way adaptive early warning system for jacking devices based on multimodal prediction. Background Technology

[0002] Lifting operations, due to their heavy loads, high-altitude environments, and complex forces, require extremely high safety standards. However, existing technologies and related patents on risk avoidance control for lifting machinery have significant shortcomings: First, the protection logic is unidirectional, often monitoring only the state of a single object such as equipment or personnel, lacking a human-machine collaborative closed loop and thus insufficient targeted protection. Second, the protection mode is reactive, relying on real-time data and fixed threshold triggers, failing to analyze predictive characteristics, and lagging behind risk evolution. Third, the perception dimension is limited, failing to cover predictive data specific to lifting scenarios, such as personnel physiological fluctuations and equipment outrigger settlement trends, and the perception modules have poor adaptability and low accuracy. Fourth, the protection strategies are rigid, often consisting of simple combinations of alarms and shutdowns, failing to dynamically adjust according to risk levels and lacking adaptability, requiring frequent manual parameter adjustments and resulting in high maintenance costs. These shortcomings prevent existing technologies from proactively mitigating core risks, easily leading to safety accidents and failing to meet the application needs of complex human-machine collaborative lifting scenarios. Summary of the Invention

[0003] Purpose of the invention: The technical problem to be solved by the present invention is to provide a human-machine bidirectional adaptive jacking device early warning system based on multimodal prediction, which addresses the shortcomings of the existing technology.

[0004] To address the aforementioned technical issues, this invention discloses a human-machine bidirectional adaptive lifting device early warning system based on multimodal prediction, comprising a predictive perception layer, a human-machine collaborative decision-making layer, and a hierarchical linkage execution layer; the architecture forms a closed-loop linkage of "predictive data acquisition - multi-dimensional trend fusion - bidirectional gradient protection - scenario-based model optimization";

[0005] The system execution includes the following steps:

[0006] Step 1: The predictive perception layer simultaneously collects predictive status trend data of personnel, which characterizes personnel risk, and fault evolution trend data of the lifting device, which reflects equipment failure.

[0007] Step 2: The personnel predictive state perception module identifies the risk of jacking device failure based on the failure evolution trend data of the jacking device; it judges the fatigue state of personnel and the risk zone accident based on the personnel predictive state trend data, and integrates the risk intention with the fatigue state of personnel to obtain the dynamic fusion result of the jacking scenario.

[0008] Step 3: The human-machine collaborative decision-making layer divides the risk level based on the dynamic fusion result of the two types of trend data in the lifting scenario, and generates a two-way protection strategy of personnel risk → device adaptation and device risk → personnel guidance.

[0009] Step 4: The hierarchical linkage execution layer executes gradient early warning and protection actions specific to the jacking operation according to the two-way protection strategy;

[0010] The predictive perception layer includes a personnel predictive state perception module and a jacking device fault evolution perception module.

[0011] The personnel predictive state perception module integrates a vibration-resistant flexible wearable sensing device, a narrow-space adaptable binocular infrared motion capture device, and a lightweight portable EEG acquisition device to simultaneously collect personnel predictive state trend data.

[0012] The predictive state trend data of personnel includes: physiological fluctuation trend data of personnel: heart rate; premonitory data of work actions: eyelid closure frequency, spatial movement paths and temporal changes of key parts of the limbs such as arms and legs, and preparatory actions for touching control keys; and attention decline trend data: electroencephalogram (EEG) signal data.

[0013] The jacking device fault evolution perception module collects jacking device fault evolution trend data through a triple path: built-in sensors, external jacking scenario customized monitoring units, and jacking control system interfaces.

[0014] The failure evolution trend data of the lifting device includes: lifting load, outrigger settlement trend, hydraulic pressure, and structural stress.

[0015] Step 2, which involves determining employee fatigue based on predictive trend data, specifically involves: pre-storing individual baseline heart rate (60-100 beats / min) and blood pressure baseline values ​​(systolic blood pressure 90-139 mmHg / diastolic blood pressure 60-89 mmHg) as individual references; and then, by real-time acquisition of eyelid closure frequency and electroencephalogram (EEG) data, determining fatigue and inattention levels.

[0016] Fatigue status: A real-time eyelid closure frequency > 50 beats / minute, lasting for ≥ 20 seconds, coupled with a real-time heart rate ≥ 10% higher than baseline, is considered fatigue. Specifically: an eyelid closure frequency of 51-55 beats / minute lasting 20-30 seconds and a real-time heart rate 10%-15% higher than baseline indicates mild fatigue; an eyelid closure frequency of 56-60 beats / minute lasting 31-60 seconds and a real-time heart rate 16%-20% higher than baseline indicates moderate fatigue; and an eyelid closure frequency > 60 beats / minute lasting > 60 seconds and a real-time heart rate > 20% higher than baseline indicates severe fatigue.

[0017] A tendency towards inattention is defined as follows: Alpha wave proportion >40% in real-time monitoring on EEG, lasting ≥30 seconds, with blood pressure fluctuation ≤5% from the individual's baseline value. Specifically: 41%-45% alpha wave proportion lasting 30-40 seconds and blood pressure fluctuation ≤5% indicates mild inattention; 46%-50% alpha wave proportion lasting 41-60 seconds and blood pressure fluctuation ≤5% indicates moderate inattention; and >50% alpha wave proportion lasting >60 seconds and blood pressure fluctuation ≤5% indicates severe inattention. The overall accuracy is ≥93%. A feature extraction model integrating basic physiological parameters (inputting baseline heart rate and blood pressure) is used to determine fatigue status (eyelid closure frequency >50 times / minute) and a tendency towards inattention (alpha wave proportion >40%), with an accuracy ≥93%.

[0018] By continuously collecting pre-operational action data, the movement direction and speed trend are fitted, and a judgment is made in conjunction with a predefined risk zone. The specific judgment logic for accidentally entering the risk zone is to meet any of the following conditions: the risk intention is judged 0.5 to 2 seconds in advance according to the following rules:

[0019] The torso or head trajectory continuously points towards the danger zone below the load, and the distance decreases by ≥5cm per frame;

[0020] The wrist trajectory points to the emergency stop external control key (distance ≤30cm);

[0021] The ankle trajectory continuously points towards the area affected by the center of gravity shift, without any tendency to turn back, indicating "entering the risk zone". Recognition accuracy ≥ 96%;

[0022] The specific steps for identifying the failure risk of the jacking device based on the failure evolution trend data of the jacking device in step 2 are as follows: The failure evolution perception module of the jacking device uses a parameter trend fitting algorithm adapted to the operating characteristics of the jacking equipment to perform continuous trend fitting and slope calculation on four core parameters in the failure evolution trend data of the jacking device: jacking load, outrigger settlement, hydraulic pressure and structural stress, to obtain the failure risk of the jacking device.

[0023] Load difference = Real-time lifting load - Rated load (real-time lifting load is the one measured earlier, and rated load is the rated load value calibrated by the lifting device). A load difference > 10% of the rated load indicates a load imbalance.

[0024] Settlement difference = Settlement of a certain outrigger within the same time window - Average settlement of all outriggers. A settlement difference greater than 2mm indicates outrigger instability.

[0025] Pressure drop rate = (initial hydraulic pressure - real-time hydraulic pressure) / time interval; a pressure drop rate > 0.5 MPa / min indicates hydraulic leakage.

[0026] Stress exceeding the allowable value by 80% constitutes structural deformation.

[0027] The parameter trend fitting algorithm is a linear least squares sliding window fitting algorithm adapted to the operating characteristics of the lifting equipment: the operating parameters of the equipment are linearly fitted through a 5-second sliding window (goodness of fit R²≥0.95), and a 3-second moving average is superimposed when extracting the hydraulic pressure drift amplitude to extract trend features such as load fluctuation rate and outrigger settlement slope.

[0028] The human-machine collaborative decision-making layer classifies risk levels into four levels: low, medium, high, and extremely dangerous. Single-dimensional scores are assessed as follows: "Normal 0 points, close to threshold 30 points, slightly exceeding limits 60 points, severely exceeding limits 100 points." Low risk (0-20 points): All personnel indicators are normal, the equipment is normal, and the load difference of the lifting device is ≤10% of the rated value, the outrigger settlement difference is ≤2mm, and the hydraulic pressure drop rate is ≤0.5MPa / min (no abnormalities such as load imbalance, outrigger instability, or hydraulic leakage). Medium risk is defined as mild personnel fatigue or slight inattention, no risky intent, and no equipment abnormalities, or normal personnel condition and a single equipment indicator close to the threshold (load difference 8%). -10% of rated value, outrigger settlement difference 1.5-2mm, hydraulic pressure drop rate 0.3-0.5MPa / min); High risk is moderate fatigue or moderate inattention of personnel with no risk intent, or mild abnormality of personnel + risk intent and no abnormality of equipment, or single indicator of equipment exceeding the limit (load difference > 10% of rated value, outrigger settlement difference > 2mm, hydraulic pressure drop rate > 0.5MPa / min) and mild abnormality of personnel, or both personnel and equipment are mildly abnormal; Extremely dangerous risk is severe fatigue or severe inattention of personnel + risk intent, or two or more indicators of equipment exceeding the limit, or severe abnormality of personnel and serious abnormality of equipment.

[0029] Each level corresponds to a differentiated lead time for prediction (0 seconds for low risk, 3 seconds for medium risk, 1.5 seconds for high risk, and 0.5 seconds for extremely dangerous risk) and protection intensity, with the weight allocation logic updated in real time according to the operation scenario.

[0030] The two-way protection strategy includes:

[0031] Personnel risk-oriented device adaptation strategy: If someone accidentally enters a risk area, an alarm will be triggered immediately and the specific risk area will be notified; for low-risk situations, only continuous monitoring is required; for medium-risk situations, the lifting device will reduce its speed by 20%~30% (reduction gradient 20% / second), and a yellow audible and visual warning will be triggered simultaneously (volume 85dB, flashing frequency 2Hz); for high-risk situations, the lifting device will reduce its speed by 50% and limit the lifting height to no more than 10% of the current height (height control accuracy ±5mm), and operation correction prompts will be pushed through AR devices; in extremely dangerous situations, the lifting device will be suspended in an emergency (braking response time ≤500ms), and the outrigger locking mechanism will be activated (linked hydraulic pressure holding circuit + mechanical locking pin, locking response ≤300ms) to achieve double safety isolation;

[0032] The personnel guidance strategy is risk-oriented: for low-risk situations, only voice prompts are given to indicate the risk source; for medium-risk situations, a dual warning is triggered by sound and light + vibration from wearable devices (vibration intensity 0.8g, lasting 3 seconds); for high-risk situations, a safe evacuation path is pushed through AR devices (overlaid with the boundary of the overturned projection area of ​​the lifting device, path deviation ≤10cm), and the lifting device load is limited to 50% of its rated value; in extremely dangerous situations, a red high-intensity sound and light warning is activated (volume 115dB, flashing frequency 5Hz), and a directional emergency evacuation command is sent (only pushed to personnel in the danger zone).

[0033] The hierarchical linkage execution layer includes a multimodal early warning unit, a lifting device dynamic adaptation unit, and a personnel precision guidance unit; the multimodal early warning unit supports audio-visual hierarchical prompts (color-coded risk level indication), wearable device vibration early warning (personalized vibration frequency matching personnel operating habits), voice directional broadcasting (zoned speaker coverage radius ≥ 20 meters) and AR visualization prompts.

[0034] The lifting device dynamic adaptation unit adopts a dual control mode of "hard-wired linkage + communication linkage". The hard-wired linkage is connected to the lifting device emergency stop circuit through a safety relay (response time ≤10ms). The communication linkage supports real-time interaction via industrial Ethernet. The emergency stop control signal has a higher priority than all operation control commands.

[0035] The personnel precision guidance unit, combined with the jacking operation spatial coordinate system (established based on the device's basic coordinates), outputs personalized evacuation paths and operation adjustment guidelines (guideline accuracy ±5cm).

[0036] The personnel precision guidance unit of the hierarchical linkage execution layer is equipped with a personalized safe evacuation path generation function. This function is based on the core operating data of the lifting device (outrigger settlement, load difference, center of gravity offset), personnel positioning data (acquired by binocular infrared motion capture equipment, with a positioning accuracy of ±5cm), basic information of the work space, and a work space coordinate system that integrates the three-dimensional terrain of the site. The specific execution steps are as follows:

[0037] Step 1: Input relevant information parameters: including safety passages within the work space (including emergency passages and temporary evacuation passages, corresponding to the core evacuation path), work area division (according to the concept of "work area" by functional zoning), real-time number and distribution of workers, algorithm assumptions and constraints (constraints include the risk weight allocation principle of "highest priority for avoiding overturning projection areas, followed by dynamic hazards, and then static obstacles," and the hard requirements of "safety redundancy distance between the path and the boundary of the overturning projection area ≥ 1.5 meters, path slope ≤ 5°, and shortest personnel evacuation time");

[0038] Step 2: Determine the number of available safe passages for evacuation: Determine the number of effective safe passages by real-time monitoring to see if they are blocked by obstacles or outside the impact range of the overturning device;

[0039] Step 3: If there is only one effective safety passage, execute the single-passage escape route selection algorithm: based on the straight-line distance between the personnel distribution location and the passage entrance, and the obstacle avoidance cost, calculate the shortest evacuation time from each area to the passage entrance, and plan a personalized path of "nearby diversion and sequential passage";

[0040] Step 4: If two or more effective safety exits exist, execute the multi-exit escape path selection algorithm, dynamically adjusting the risk level in conjunction with the human-machine collaborative decision-making layer. Specifically, this includes:

[0041] Preliminary allocation based on proximity: Based on personnel positioning data collected by binocular infrared motion capture equipment (positioning accuracy ±5cm), the straight-line distance from personnel in each area to the entrance of each effective passage is calculated (ignoring minor obstacles and based on the shortest path), and personnel are preferentially allocated to the nearest effective passage.

[0042] Risk-adaptive batch division: The number of people evacuated in each batch is dynamically determined according to the risk level of the human-machine collaborative decision-making layer: ≤20 people per batch in low / medium risk; ≤15 people per batch in high / extremely high risk (to avoid crowding and congestion in a single channel); the interval between adjacent batches in the same channel is fixed at 1 second (shortened to 0.5 seconds in high / extremely high risk to speed up evacuation efficiency), and “batch number + release instruction” is synchronized through AR glasses and voice prompts.

[0043] Dynamic adaptation and adjustment: If a passage is suddenly blocked (e.g., due to new obstacles or entry into the area affected by a device overturning), immediately reassign personnel waiting to be assigned to that passage to the next nearest available safe passage, with an adjustment response time ≤ 500ms; if the cumulative occupancy rate of a passage is ≥ 70% (≥ 60% in high / extremely dangerous situations), suspend new batch assignments for that passage and transfer subsequent personnel to other passages to avoid passage overload.

[0044] Step 5: Dynamically verify information changes: If the number of personnel increases or decreases, personnel positions move, the status of safety passages changes (such as new obstacles or passages becoming available again) or the danger zone of the device expands or shrinks during the operation, return to Step 1 to re-enter the updated parameters and re-execute the path planning;

[0045] Step 6: Path Output and Guidance: When there are no changes in information, the boundary of the dynamic tilt projection area and the safe evacuation path are output in real time through the smart terminal at the work site and industrial-grade AR glasses. Dangerous restricted areas and safe passage routes are clearly marked, achieving personalized safety guidance with path deviation ≤10cm and guidance accuracy ±5cm. Among them, AR visualization prompts support dynamic updates of evacuation paths, channel congestion warnings, and optimal route adjustment prompts.

[0046] In the process of generating the personalized safe evacuation path, a rigid body dynamics simulation model is used. The rated overturning moment of the jacking device, the current jacking height, and real-time operation data are input to generate a dynamic overturning projection area boundary with an update frequency of ≥10Hz and a boundary positioning accuracy of ±3cm. The coordinate information of static obstacles, dynamic danger points, and preset safety assembly points at the work site is superimposed to construct a full-scene risk map, providing basic data support for the path selection algorithm.

[0047] The AR visualization prompts are achieved through industrial-grade AR glasses with a resolution ≥1920×1080, a field of view ≥50°, and a brightness ≥500cd / m² (visible in sunlight). It supports visualization of the boundary of the lifting danger zone, prediction of the equipment's movement trajectory (predicting the trajectory for the next 2 seconds based on current operating parameters), and prompts for correcting operational errors. The dynamic adaptation unit of the lifting device is compatible with ModbusRTU, Profinet, and EtherNet / IP industrial communication protocols, with a control command transmission delay ≤100ms. It can adapt to hydraulic (single-cylinder / multi-cylinder synchronous), mechanical (screw / gear transmission), and electric (screw / winch drive) lifting devices, covering ≥90% of mainstream lifting equipment types.

[0048] The system adopts a modular design, which can be quickly adapted to various human-machine collaborative lifting scenarios such as equipment installation lifting, building component lifting, construction machinery lifting, and high-altitude operation lifting by replacing the dedicated sensing unit for the lifting scenario (laser displacement sensor for outrigger settlement, piezoelectric patch sensor for structural stress), adjusting trend analysis parameters (fluctuation rate threshold under different load levels), and configuring protection strategy templates (preset protection logic for multiple scenarios). The module replacement time is ≤30 minutes.

[0049] Beneficial effects:

[0050] The present invention provides a method, system, and lifting device based on multimodal prediction for human-machine bidirectional adaptive early warning and protection. Through a predictive perception layer, it synchronously collects predictive trend data of personnel and equipment, providing comprehensive support for human-machine collaborative decision-making. The human-machine collaborative decision-making layer accurately determines risk levels based on a dynamic weight fusion strategy, generating bidirectional protection strategies. The hierarchical linkage execution layer provides differentiated protection according to risk levels, balancing safety and efficiency. The personnel guidance function, through personalized path planning, AR visualization guidance, and multimodal early warning, quickly responds to high-risk evacuation needs, ensuring safe and efficient personnel evacuation. Attached Figure Description

[0051] Figure 1 is a flowchart of the architecture of the human-machine two-way adaptive early warning and protection system.

[0052] Figure 2 is a flowchart of the human-machine two-way early warning and protection system.

[0053] Figure 3 shows a simulation of the safe evacuation of personnel from the lifting device in a high-risk scenario. Detailed Implementation

[0054] Lifting devices, as core equipment for heavy-duty lifting, are widely used in scenarios such as building component installation, construction machinery maintenance, and aerial work platforms. The safety of lifting operations is directly related to the safety of personnel and equipment property. The risks mainly stem from two aspects: first, misjudgment of personnel's operational intentions and violations of regulations caused by fluctuations in their mental state; second, sudden accidents caused by the evolution of faults such as equipment outrigger instability, hydraulic leakage, and excessive structural stress. Existing protection technologies are mostly one-way response protection, relying solely on real-time data to trigger alarms or shutdowns. They lack the ability to predict personnel status trends and early signs of equipment failure, and their perception dimensions are limited, with poor scenario adaptability, making it difficult to meet the safety requirements of complex human-machine collaborative lifting scenarios.

[0055] To address the aforementioned technical problems, this invention discloses a human-machine bidirectional adaptive lifting device early warning system based on multimodal prediction. The system is characterized by comprising a predictive perception layer, a human-machine collaborative decision-making layer, and a hierarchical linkage execution layer; the architecture forms a closed-loop linkage of "predictive data acquisition - multi-dimensional trend fusion - bidirectional gradient protection - scenario-based model optimization."

[0056] The system execution includes the following steps:

[0057] Step 1: The predictive perception layer simultaneously collects predictive status trend data of personnel, which characterizes personnel risk, and fault evolution trend data of the lifting device, which reflects equipment failure.

[0058] Step 2: The personnel predictive state perception module identifies the risk of jacking device failure based on the failure evolution trend data of the jacking device; it judges the fatigue state of personnel and the risk zone accident based on the personnel predictive state trend data, and integrates the risk intention with the fatigue state of personnel to obtain the dynamic fusion result of the jacking scenario.

[0059] Step 3: The human-machine collaborative decision-making layer divides the risk level based on the dynamic fusion result of the two types of trend data in the lifting scenario, and generates a two-way protection strategy of personnel risk → device adaptation and device risk → personnel guidance.

[0060] Step 4: The hierarchical linkage execution layer executes gradient early warning and protection actions specific to the jacking operation according to the two-way protection strategy.

[0061] This embodiment uses hydraulic jacking of building components as an application scenario. The component weighs 50t and the working space is 10m long, 8m wide and 6m high. In practical applications, this method can also be adapted to equipment installation jacking, construction machinery jacking, high-altitude operation jacking and other scenarios. The corresponding jacking device types can include hydraulic, mechanical and electric types, etc., but this invention is not limited to these.

[0062] For example, hardware selection needs to be specifically adapted to the jacking operation environment: For personnel sensing, a flexible wearable sensing device with an IP68 protection rating, a sampling frequency of 10Hz, and a heart rate measurement accuracy of ±1 beat / minute is selected; a binocular infrared motion capture device with a resolution of 2560×1440 is used, installed at a height of 4m, with a coverage radius ≥15m; an 8-channel portable EEG acquisition device is used, with a sampling rate of 250Hz. For equipment sensing, vibration sensors are deployed at the base of the outriggers, laser displacement sensors are used for outrigger settlement monitoring, with an accuracy of ±0.01mm; stress sensors are attached to key stress points of the jacking beam, and acoustic sensors are deployed on the hydraulic pump group. For decision-making and execution, an edge computing gateway with a computing power ≥8 TOPS is configured, along with industrial-grade AR glasses with a resolution of 1920×1080 and a brightness of 500cd / m²; a color-coded audible and visual alarm with a volume of 85~115dB; and a safety relay with a response time ≤10ms.

[0063] Parameter initialization requires the input of basic scenario data, equipment parameters, and personnel information: Scenario parameters include component weight, workspace coordinates, outrigger spacing, etc.; equipment parameters include rated load of the lifting device, rated pressure of the hydraulic system, lifting speed, etc.; personnel information includes basic physiological parameters of the operators, operator proficiency level, etc. Simultaneously, dynamic initial weight values ​​are configured according to the characteristics of the lifting scenario. In heavy-load scenarios (load ≥ 80% of rated value), the weights for both personnel status trends and equipment failure evolution trends are set to 50%, and risk level thresholds are defined: low risk is no trend exceeding the threshold, medium risk is one trend approaching the threshold, high risk is two trends exceeding the threshold, and extremely high risk is three or more trends exceeding the threshold.

[0064] Communication debugging requires establishing industrial Ethernet communication between the predictive perception layer, the human-machine collaborative decision-making layer, and the hierarchical linkage execution layer, supporting the EtherNet / IP protocol, with a transmission latency of ≤300ms, a test hard-wired emergency stop circuit response time of ≤500ms, and AR glasses data transmission latency of ≤50ms, to ensure reliable linkage of each module.

[0065] The predictive perception layer includes a personnel predictive state perception module and a jacking device fault evolution perception module.

[0066] The personnel predictive state perception module integrates a vibration-resistant flexible wearable sensing device, a narrow-space adaptable binocular infrared motion capture device, and a lightweight portable EEG acquisition device to simultaneously collect personnel predictive state trend data.

[0067] The predictive state trend data of personnel includes: physiological fluctuation trend data of personnel: heart rate; premonitory data of work actions: eyelid closure frequency, spatial movement paths and temporal changes of key parts of the limbs such as arms and legs, and preparatory actions for touching control keys; and attention decline trend data: electroencephalogram (EEG) signal data.

[0068] The jacking device fault evolution perception module collects jacking device fault evolution trend data through a triple path: built-in sensors, external jacking scenario customized monitoring units, and jacking control system interfaces.

[0069] The failure evolution trend data of the lifting device includes: lifting load, outrigger settlement trend, hydraulic pressure, and structural stress.

[0070] Step 2, which involves determining employee fatigue based on predictive trend data, specifically involves: pre-storing individual baseline heart rate (60-100 beats / min) and blood pressure baseline values ​​(systolic blood pressure 90-139 mmHg / diastolic blood pressure 60-89 mmHg) as individual references; and then, by real-time acquisition of eyelid closure frequency and electroencephalogram (EEG) data, determining fatigue and inattention levels.

[0071] Fatigue status: A real-time eyelid closure frequency > 50 beats / minute, lasting for ≥ 20 seconds, coupled with a real-time heart rate ≥ 10% higher than baseline, is considered fatigue. Specifically: an eyelid closure frequency of 51-55 beats / minute lasting 20-30 seconds and a real-time heart rate 10%-15% higher than baseline indicates mild fatigue; an eyelid closure frequency of 56-60 beats / minute lasting 31-60 seconds and a real-time heart rate 16%-20% higher than baseline indicates moderate fatigue; and an eyelid closure frequency > 60 beats / minute lasting > 60 seconds and a real-time heart rate > 20% higher than baseline indicates severe fatigue.

[0072] Inattention is defined as follows: A real-time monitoring value of alpha waves on an EEG >40% for ≥30 seconds, accompanied by blood pressure fluctuations ≤5% from the individual's baseline value. Specifically: 41%-45% alpha waves for 30-40 seconds and blood pressure fluctuations ≤5% indicate mild inattention; 46%-50% for 41-60 seconds and blood pressure fluctuations ≤5% indicate moderate inattention; and >50% for >60 seconds and blood pressure fluctuations ≤5% indicate severe inattention. The overall accuracy is ≥93%. A feature extraction model incorporating basic physiological parameters (inputting baseline heart rate and blood pressure) is used to assess fatigue (eyelid closure frequency >50 times / minute) and inattention trends (alpha wave percentage >40%), with an accuracy ≥93%.

[0073] By continuously collecting pre-operational action data, the movement direction and speed trend are fitted, and a judgment is made in conjunction with a predefined risk zone. The specific judgment logic for accidentally entering the risk zone is to meet any of the following conditions: the risk intention is judged 0.5 to 2 seconds in advance according to the following rules:

[0074] The torso or head trajectory continuously points towards the danger zone below the load, and the distance decreases by ≥5cm per frame;

[0075] The wrist trajectory points to the emergency stop external control key (distance ≤30cm);

[0076] The ankle trajectory continuously points towards the area affected by the center of gravity shift, without any tendency to turn back, indicating "entering the risk zone". Recognition accuracy ≥ 96%;

[0077] The specific steps for identifying the failure risk of the jacking device based on the failure evolution trend data of the jacking device in step 2 are as follows: The failure evolution perception module of the jacking device uses a parameter trend fitting algorithm adapted to the operating characteristics of the jacking equipment to perform continuous trend fitting and slope calculation on four core parameters in the failure evolution trend data of the jacking device: jacking load, outrigger settlement, hydraulic pressure and structural stress, to obtain the failure risk of the jacking device.

[0078] Load difference = Real-time lifting load - Rated load (real-time lifting load is the one measured earlier, and rated load is the rated load value calibrated by the lifting device). A load difference > 10% of the rated load indicates a load imbalance.

[0079] Settlement difference = Settlement of a certain outrigger within the same time window - Average settlement of all outriggers. A settlement difference greater than 2mm indicates outrigger instability.

[0080] Pressure drop rate = (initial hydraulic pressure - real-time hydraulic pressure) / time interval; a pressure drop rate > 0.5 MPa / min indicates hydraulic leakage.

[0081] Stress exceeding the allowable value by 80% constitutes structural deformation.

[0082] The parameter trend fitting algorithm is a linear least squares sliding window fitting algorithm adapted to the operating characteristics of the lifting equipment: the operating parameters of the equipment are linearly fitted through a 5-second sliding window (goodness of fit R²≥0.95), and a 3-second moving average is superimposed when extracting the hydraulic pressure drift amplitude to extract trend features such as load fluctuation rate and outrigger settlement slope.

[0083] The human-machine collaborative decision-making layer classifies risk levels into four levels: low, medium, high, and extremely dangerous. Single-dimensional scores are assessed as follows: "Normal 0 points, close to threshold 30 points, slightly exceeding limits 60 points, severely exceeding limits 100 points." Low risk (0-20 points): All personnel indicators are normal, the equipment is normal, and the load difference of the lifting device is ≤10% of the rated value, the outrigger settlement difference is ≤2mm, and the hydraulic pressure drop rate is ≤0.5MPa / min (no abnormalities such as load imbalance, outrigger instability, or hydraulic leakage). Medium risk is defined as mild personnel fatigue or slight inattention, no risky intent, and no equipment abnormalities, or normal personnel condition and a single equipment indicator close to the threshold (load difference 8%). -10% of rated value, outrigger settlement difference 1.5-2mm, hydraulic pressure drop rate 0.3-0.5MPa / min); High risk is moderate fatigue or moderate inattention of personnel with no risk intent, or mild abnormality of personnel + risk intent and no abnormality of equipment, or single indicator of equipment exceeding the limit (load difference > 10% of rated value, outrigger settlement difference > 2mm, hydraulic pressure drop rate > 0.5MPa / min) and mild abnormality of personnel, or both personnel and equipment are mildly abnormal; Extremely dangerous risk is severe fatigue or severe inattention of personnel + risk intent, or two or more indicators of equipment exceeding the limit, or severe abnormality of personnel and serious abnormality of equipment.

[0084] Each level corresponds to a differentiated lead time for prediction (0 seconds for low risk, 3 seconds for medium risk, 1.5 seconds for high risk, and 0.5 seconds for extremely dangerous risk) and protection intensity, with the weight allocation logic updated in real time according to the operation scenario.

[0085] The two-way protection strategy includes:

[0086] Personnel risk-oriented device adaptation strategy: If someone accidentally enters a risk area, an alarm will be triggered immediately and the specific risk area will be notified; for low-risk situations, only continuous monitoring is required; for medium-risk situations, the lifting device will reduce its speed by 20%~30% (reduction gradient 20% / second), and a yellow audible and visual warning will be triggered simultaneously (volume 85dB, flashing frequency 2Hz); for high-risk situations, the lifting device will reduce its speed by 50% and limit the lifting height to no more than 10% of the current height (height control accuracy ±5mm), and operation correction prompts will be pushed through AR devices; in extremely dangerous situations, the lifting device will be suspended in an emergency (braking response time ≤500ms), and the outrigger locking mechanism will be activated (linked hydraulic pressure holding circuit + mechanical locking pin, locking response ≤300ms) to achieve double safety isolation;

[0087] The personnel guidance strategy is risk-oriented: for low-risk situations, only voice prompts are given to indicate the risk source; for medium-risk situations, a dual warning is triggered by sound and light + vibration from wearable devices (vibration intensity 0.8g, lasting 3 seconds); for high-risk situations, a safe evacuation path is pushed through AR devices (overlaid with the boundary of the overturned projection area of ​​the lifting device, path deviation ≤10cm), and the lifting device load is limited to 50% of its rated value; in extremely dangerous situations, a red high-intensity sound and light warning is activated (volume 115dB, flashing frequency 5Hz), and a directional emergency evacuation command is sent (only pushed to personnel in the danger zone).

[0088] The hierarchical linkage execution layer includes a multimodal early warning unit, a lifting device dynamic adaptation unit, and a personnel precision guidance unit; the multimodal early warning unit supports audio-visual hierarchical prompts (color-coded risk level indication), wearable device vibration early warning (personalized vibration frequency matching personnel operating habits), voice directional broadcasting (zoned speaker coverage radius ≥ 20 meters) and AR visualization prompts.

[0089] The lifting device dynamic adaptation unit adopts a dual control mode of "hard-wired linkage + communication linkage". The hard-wired linkage is connected to the lifting device emergency stop circuit through a safety relay (response time ≤10ms). The communication linkage supports real-time interaction via industrial Ethernet. The emergency stop control signal has a higher priority than all operation control commands.

[0090] The personnel precision guidance unit, combined with the jacking operation spatial coordinate system (established based on the device's basic coordinates), outputs personalized evacuation paths and operation adjustment guidelines (guideline accuracy ±5cm).

[0091] The personnel precision guidance unit of the hierarchical linkage execution layer is equipped with a personalized safe evacuation path generation function. This function is based on the core operating data of the lifting device (outrigger settlement, load difference, center of gravity offset), personnel positioning data (acquired by binocular infrared motion capture equipment, with a positioning accuracy of ±5cm), basic information of the work space, and a work space coordinate system that integrates the three-dimensional terrain of the site. The specific execution steps are as follows:

[0092] Step 1: Input relevant information parameters: including safety passages within the work space (including emergency passages and temporary evacuation passages, corresponding to the core evacuation path), work area division (according to the concept of "work area" by functional zoning), real-time number and distribution of workers, algorithm assumptions and constraints (constraints include the risk weight allocation principle of "highest priority for avoiding overturning projection areas, followed by dynamic hazards, and then static obstacles," and the hard requirements of "safety redundancy distance between the path and the boundary of the overturning projection area ≥ 1.5 meters, path slope ≤ 5°, and shortest personnel evacuation time");

[0093] Step 2: Determine the number of available safe passages for evacuation: Determine the number of effective safe passages by real-time monitoring to see if they are blocked by obstacles or outside the impact range of the overturning device;

[0094] Step 3: If there is only one effective safety passage, execute the single-passage escape route selection algorithm: based on the straight-line distance between the personnel distribution location and the passage entrance, and the obstacle avoidance cost, calculate the shortest evacuation time from each area to the passage entrance, and plan a personalized path of "nearby diversion and sequential passage";

[0095] Step 4: If two or more effective safety exits exist, execute the multi-exit escape path selection algorithm, dynamically adjusting the risk level in conjunction with the human-machine collaborative decision-making layer. Specifically, this includes:

[0096] Preliminary allocation based on proximity: Based on personnel positioning data collected by binocular infrared motion capture equipment (positioning accuracy ±5cm), the straight-line distance from personnel in each area to the entrance of each effective passage is calculated (ignoring minor obstacles and based on the shortest path), and personnel are preferentially allocated to the nearest effective passage.

[0097] Risk-adaptive batch division: The number of people evacuated in each batch is dynamically determined according to the risk level of the human-machine collaborative decision-making layer: ≤20 people per batch in low / medium risk; ≤15 people per batch in high / extremely high risk (to avoid crowding and congestion in a single channel); the interval between adjacent batches in the same channel is fixed at 1 second (shortened to 0.5 seconds in high / extremely high risk to speed up evacuation efficiency), and “batch number + release instruction” is synchronized through AR glasses and voice prompts.

[0098] Dynamic adaptation and adjustment: If a passage is suddenly blocked (e.g., due to new obstacles or entry into the area affected by a device overturning), immediately reassign personnel waiting to be assigned to that passage to the next nearest available safe passage, with an adjustment response time ≤ 500ms; if the cumulative occupancy rate of a passage is ≥ 70% (≥ 60% in high / extremely dangerous situations), suspend new batch assignments for that passage and transfer subsequent personnel to other passages to avoid passage overload.

[0099] Step 5: Dynamically verify information changes: If the number of personnel increases or decreases, personnel positions move, the status of safety passages changes (such as new obstacles or passages becoming available again) or the danger zone of the device expands or shrinks during the operation, return to Step 1 to re-enter the updated parameters and re-execute the path planning;

[0100] Step 6: Path Output and Guidance: When there are no changes in information, the boundary of the dynamic tilt projection area and the safe evacuation path are output in real time through the smart terminal at the work site and industrial-grade AR glasses. Dangerous restricted areas and safe passage routes are clearly marked, achieving personalized safety guidance with path deviation ≤10cm and guidance accuracy ±5cm. Among them, AR visualization prompts support dynamic updates of evacuation paths, channel congestion warnings, and optimal route adjustment prompts.

[0101] In the process of generating the personalized safe evacuation path, a rigid body dynamics simulation model is used. The rated overturning moment of the jacking device, the current jacking height, and real-time operation data are input to generate a dynamic overturning projection area boundary with an update frequency of ≥10Hz and a boundary positioning accuracy of ±3cm. The coordinate information of static obstacles, dynamic danger points, and preset safety assembly points at the work site is superimposed to construct a full-scene risk map, providing basic data support for the path selection algorithm.

[0102] The AR visualization prompts are achieved through industrial-grade AR glasses with a resolution ≥1920×1080, a field of view ≥50°, and a brightness ≥500cd / m² (visible in sunlight). It supports visualization of the boundary of the lifting danger zone, prediction of the equipment's movement trajectory (predicting the trajectory for the next 2 seconds based on current operating parameters), and prompts for correcting operational errors. The dynamic adaptation unit of the lifting device is compatible with ModbusRTU, Profinet, and EtherNet / IP industrial communication protocols, with a control command transmission delay ≤100ms. It can adapt to hydraulic (single-cylinder / multi-cylinder synchronous), mechanical (screw / gear transmission), and electric (screw / winch drive) lifting devices, covering ≥90% of mainstream lifting equipment types.

[0103] The system adopts a modular design, which can be quickly adapted to various human-machine collaborative lifting scenarios such as equipment installation lifting, building component lifting, construction machinery lifting, and high-altitude operation lifting by replacing the dedicated sensing unit for the lifting scenario (laser displacement sensor for outrigger settlement, piezoelectric patch sensor for structural stress), adjusting trend analysis parameters (fluctuation rate threshold under different load levels), and configuring protection strategy templates (preset protection logic for multiple scenarios). The module replacement time is ≤30 minutes.

[0104] Application Examples and Implementation Results

[0105] The following section details the practical application process of this invention using a hydraulic jacking operation scenario for building components:

[0106] At a construction site, a 4-cylinder hydraulic jacking device was used to lift a 50-ton concrete component. The working space was 10m long, 8m wide, and 6m high, with 3 operators working together. Two safety passages were pre-set within the working space (one on the east side and one on the west side, both emergency passages, free of temporary obstacles, with a safety redundancy distance of ≥1.5 meters from the boundary of the overturning projection area), as shown in Figure 3. After system deployment and initialization, scene parameters (including safety passage locations, work area division, and basic physiological data of personnel) were entered, hardware debugging and communication link testing were performed (industrial Ethernet transmission latency ≤280ms, AR glasses data transmission latency ≤45ms), and then the system entered the normal lifting phase.

[0107] Within the initial 0 to 1.5 hours, the sensor layer monitoring showed that the personnel's physiological state was stable, the heart rate was maintained at 70±5 beats / minute, the alpha wave ratio was 30%, the binocular infrared motion capture device did not capture any premonitory movements pointing towards the risk area, the device parameters were normal, the outrigger subsidence was 0.02mm / s, the hydraulic pressure was 20MPa, and the decision layer determined that the risk was low and continued monitoring was carried out.

[0108] At 1.5 hours, the risk evolution stage begins: the auxiliary observation hand becomes fatigued due to continuous operation, the wearable device shows that the heart rate variability fluctuates by ±25%, and the proportion of alpha waves in the brain increases to 45%; wear of the hydraulic pump group causes the outrigger to sink faster, with a slope of 0.12 mm / s, and the acoustic sensor detects abnormal noise. After the perception layer preprocesses the data and uploads it, the decision layer determines within 1.5 seconds that there is a high risk of outrigger instability and generates a two-way protection strategy: the lifting device is reduced by 50% and the height is limited to 3.3m. At the same time, a red sound and light warning and wearable vibration prompt are activated, the AR glasses push the outrigger settlement acceleration, the personnel precision guidance unit is activated, and a personalized safe evacuation path is generated. The input includes the real-time status of two safety passages: both are unobstructed and outside the impact range of the device overturning. The real-time distribution of three workers is as follows: the main operator is located in the control console area, and the auxiliary observers A and B are located in the south and north working areas of the component, respectively. After the system inputs updated parameters containing the constraints of "overturning projection area avoidance priority is the highest and path slope ≤ 5°", it determines that the two safety passages are valid and executes the multi-channel escape path selection algorithm. Based on the binocular infrared motion capture data with a positioning accuracy of ±4cm, A, who is 3.2m away from the east passage, is given priority to be assigned to the east passage, and B, who is 2.8m away from the west passage, is given priority to be assigned to the west passage. The main operator does not evacuate temporarily because he needs to perform a deceleration operation. Given the high-risk level, following the rule of ≤15 people per batch and 0.5-second interval between adjacent batches in the passage, the AR glasses simultaneously issued the "1 batch + immediate evacuation" command. During the evacuation, the system confirmed that there were no obstacles in the passage and that the personnel's trajectory was normal. It then pushed the eastern passage path with the superimposed dynamic tilting projection area boundary (update frequency 12Hz, positioning accuracy ±2cm) and path deviation ≤8cm to person A, and pushed the western passage path marked "avoiding the impact area of ​​the south side center of gravity shift of the component" to person B. At the same time, the system issued a voice warning through the zoned speakers with a coverage radius of 25 meters, activated a yellow sound and light warning with a volume of 95dB and a flashing frequency of 3Hz, and activated a wearable device vibration warning with a vibration intensity of 0.8g and a duration of 3 seconds. Two auxiliary observers safely arrived at the assembly point within 12 seconds following the instructions.

[0109] After the operator evacuated the danger zone following AR guidance, the main operator inspected the hydraulic pump unit and replaced the seals. The outrigger settlement returned to 0.02 mm / s. The edge node uploaded the event data, and the cloud optimized the model parameters: the attention threshold for novice operators was relaxed to 45% alpha wave proportion, and the outrigger settlement warning slope for the aging device was adjusted to 0.12 mm / s. The system then loaded the optimized parameters over the next 1.5 hours, successfully completing the 5m lifting target without triggering any further high-risk warnings.

[0110] Implementation results show that the system has a 97% accuracy rate in identifying personnel risk intent, a 95% accuracy rate in predicting equipment failure, a ±4cm accuracy in guiding personnel evacuation in high-risk scenarios, a path deviation of ≤8cm, an 18% improvement in operational efficiency compared to existing systems, a 30-minute reduction in excessive downtime, a 35% reduction in maintenance costs, and requires no manual parameter adjustments, fully verifying the feasibility and advancement of the technical solution.

[0111] This invention provides a human-machine bidirectional adaptive jacking device early warning system based on multimodal prediction. Many methods and approaches exist for implementing this technical solution; the above description is merely a preferred embodiment of the invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this invention, and these improvements and modifications should also be considered within the scope of protection of this invention. All components not explicitly stated in this embodiment can be implemented using existing technologies.

Claims

1. A human-machine bidirectional adaptive jacking device early warning system based on multimodal prediction, characterized in that, The system comprises a predictive perception layer, a human-machine collaborative decision-making layer, and a hierarchical linkage execution layer. The execution of this system includes the following steps: Step 1: The predictive perception layer synchronously collects predictive state trend data representing personnel risk and jacking device failure evolution trend data reflecting equipment failure; Step 2: The personnel predictive state perception module identifies jacking device failure risks based on the jacking device failure evolution trend data; and judges personnel fatigue status and accidental entry into risk areas based on the personnel predictive state trend data; Step 3: The human-machine collaborative decision-making layer classifies risk levels based on the dynamic fusion results of the two types of trend data in the jacking scenario, generating a two-way protection strategy of personnel risk → device adaptation and device risk → personnel guidance; Step 4: The hierarchical linkage execution layer executes jacking operation-specific gradient early warning and protection actions, as well as personalized safe evacuation path generation and guidance, according to the two-way protection strategy.

2. The early warning system for a human-machine bidirectional adaptive lifting device based on multimodal prediction as described in claim 1, characterized in that, The predictive perception layer includes a personnel predictive state perception module and a jacking device fault evolution perception module. The predictive state perception module collects predictive state trend data including: physiological fluctuation trend data: heart rate; premonitory data of work actions: eyelid closure frequency, spatial movement path and timing changes of key limbs, and preparatory actions for touch control keys; and attention decay trend data: electroencephalogram (EEG) signal data.

3. The early warning system for a human-machine bidirectional adaptive lifting device based on multimodal prediction according to claim 2, characterized in that, The fault evolution perception module of the lifting device collects fault evolution trend data of the lifting device, including: lifting load, outrigger settlement trend, hydraulic pressure and structural stress.

4. The early warning system for a human-machine bidirectional adaptive lifting device based on multimodal prediction according to claim 2, characterized in that, The step 2, which involves judging the fatigue state of personnel based on predictive state trend data, specifically involves: pre-storing the individual's baseline heart rate and blood pressure as individual references, and using real-time collected eyelid closure frequency and EEG signal data to determine fatigue state and inattention state; the judgment of accidentally entering a risk area specifically involves: fitting the movement direction and speed trend through continuously collected pre-action data, and making a judgment based on predefined risk areas.

5. The early warning system for a human-machine bidirectional adaptive lifting device based on multimodal prediction according to claim 2, characterized in that, The step 2, which identifies the failure risk of the lifting device based on the failure evolution trend data of the lifting device, specifically involves: performing continuous trend fitting and slope calculation on the failure evolution trend data of the lifting device using a parameter trend fitting algorithm to obtain the failure risk of the lifting device.

6. The early warning system for a human-machine bidirectional adaptive lifting device based on multimodal prediction according to claim 5, characterized in that, The parameter trend fitting algorithm is a linear least squares sliding window fitting algorithm adapted to the operating characteristics of the lifting equipment: the operating parameters of the equipment are linearly fitted through an n1-second sliding window, and the hydraulic pressure drift amplitude is extracted by superimposing an n2-second moving average to extract trend features including load fluctuation rate and outrigger settlement slope.

7. The early warning system for a human-machine bidirectional adaptive lifting device based on multimodal prediction according to claim 1, characterized in that, The human-machine collaborative decision-making layer classifies the risk level into four levels: low, medium, high, and extremely dangerous, based on the combined fatigue state and the risk of jacking device failure. Each level corresponds to a differentiated lead time for prediction and protection intensity, and the weight allocation logic is updated in real time according to the operation scenario.

8. The early warning system for a human-machine bidirectional adaptive lifting device based on multimodal prediction according to claim 1, characterized in that, The hierarchical linkage execution layer includes a multimodal early warning unit, a lifting device dynamic adaptation unit, and a personnel precision guidance unit. The multimodal early warning unit supports tiered sound and light prompts, wearable device vibration warnings, voice directional broadcasts, and AR visual prompts for precise personnel evacuation path planning and output. The lifting device dynamic adaptation unit adopts a dual control mode of "hard-wired linkage + communication linkage." The hard-wired linkage is connected to the lifting device's emergency stop circuit via a safety relay, while the communication linkage supports real-time interaction via industrial Ethernet. The emergency stop control signal has higher priority than all operation control commands. The personnel precision guidance unit, combined with the lifting operation spatial coordinate system, outputs personalized evacuation paths and operation adjustment guidelines.

9. The early warning system for a human-machine bidirectional adaptive lifting device based on multimodal prediction according to claim 8, characterized in that, The two-way protection strategy includes: if someone accidentally enters a risk area, an alarm will be triggered immediately and the specific risk area will be notified; for low-risk situations, only continuous monitoring will be conducted; for medium-risk situations, the lifting device will slow down and a yellow audible and visual warning will be triggered simultaneously; for high-risk situations, a precise evacuation route will be pushed through AR devices; and in extremely dangerous situations, a red high-intensity audible and visual warning will be activated to send directional emergency evacuation instructions to personnel in the danger zone and push the optimal evacuation route updated in real time.

10. The early warning system for a human-machine bidirectional adaptive lifting device based on multimodal prediction according to claim 8, characterized in that, The generation of the precise evacuation path is based on the core operating data of the lifting device, personnel positioning data, basic information of the work space, and a work space coordinate system that integrates the three-dimensional terrain of the site. The specific execution steps are as follows: Step 1.1: Input relevant information parameters including safety passages within the work space, work area division, real-time number and distribution of personnel, algorithm assumptions, and constraints; Step 2.1: Determine the number of available safety passages for evacuation; Step 3.1: If there is only one effective safety passage, execute the single-passage escape path selection algorithm: calculate the shortest evacuation time from the personnel distribution location to the passage entrance and the obstacle avoidance cost, and plan the path; Step 4.1: If there are two or more effective safety passages, execute the multi-passage escape path selection algorithm, and plan the path by combining the risk level dynamic adjustment of the human-machine collaborative decision-making layer. The risk level dynamic adjustment specifically includes: preliminary proximity allocation, risk adaptation batch division, and dynamic adaptation adjustment; Step 5.1: Dynamically verify information changes: if there is an increase or decrease in the number of personnel, movement of personnel positions, changes in the status of safety passages, or expansion of the danger range of the device during the operation... Zoom out, return to step 1 to re-enter the updated parameters, and re-execute path planning; Step 6.1: Path output and guidance.

Images