Amusement facility driving safety intelligent evaluation system based on artificial intelligence

Abstract

The invention relates to the technical field of safety evaluation, in particular to a recreation facility driving safety intelligent evaluation system based on artificial intelligence. Comprising a data acquisition module, an intelligent evaluation platform and a visual terminal, the intelligent evaluation platform is used for performing feature extraction and modeling on the collected operation parameters, and calling an artificial intelligence algorithm to generate a safety level evaluation result corresponding to the driving state; the artificial intelligence algorithm comprises a state recognition model based on deep learning and a risk prediction model based on probability inference, the state recognition model is used for judging the operation mode of the driving mechanism, and the risk prediction model is used for predicting the occurrence probability of potential faults. The method comprises the following steps: constructing a data acquisition module to acquire operation characteristic parameters of a driving device; introducing an artificial intelligence model, and performing feature extraction and anomaly detection on the data; based on the historical operation data and the model, generating a security level and predicting a potential risk; and an evaluation result is transmitted to a visual terminal, so that a manager can make a decision immediately.

Description

Technical Field

[0001] This invention relates to the field of safety assessment technology, specifically to an artificial intelligence-based intelligent safety assessment system for amusement facilities. Background Technology

[0002] With the increasing prevalence of large-scale amusement rides in the modern entertainment industry, their drive mechanisms, as the core component of the entire system, bear the critical responsibilities of power output and safety protection. However, most facilities currently still employ traditional detection methods, primarily relying on periodic manual inspections and simple alarm mechanisms based on fixed thresholds. These methods are not only inefficient and highly subjective, but also struggle to capture and comprehensively analyze the multi-state parameters of equipment under complex operating environments in real time.

[0003] In actual operation, key components such as drive motors, gearboxes, and transmission chains are often under high-frequency and high-load conditions, which can easily lead to potential risks due to long-term fatigue accumulation, material aging, or sudden overload. If these damages and abnormal conditions are not detected and diagnosed in time, they are very likely to gradually evolve during operation, eventually causing malfunctions or even serious safety accidents, threatening the lives and property of tourists.

[0004] Therefore, given the rapid development of intelligent technologies, there is an urgent need to introduce AI-driven health assessment systems into equipment operation and maintenance management. This system should be able to dynamically monitor, extract features, and classify safety levels of drive devices based on real-time sensor data using machine learning, deep learning, and other methods, thus shifting from "post-incident handling" to "pre-incident early warning." This will not only significantly improve the overall safety of large-scale amusement facilities but also greatly enhance the efficiency and accuracy of regulatory work, providing key technological support for the sustainable development of the industry. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent assessment system for the safety of amusement rides based on artificial intelligence, which solves the problem that existing monitoring methods are unable to detect potential hazards in a timely and accurate manner.

[0006] This invention constructs a data acquisition module to collect the operating characteristic parameters of the driving device; introduces an artificial intelligence model to extract features and detect anomalies in the data; generates a safety level and predicts potential risks based on historical operating data and the model; and transmits the evaluation results to a visualization terminal so that managers can make decisions in real time.

[0007] To achieve the above objectives, the present invention provides the following technical solution: It includes a data acquisition module, an intelligent evaluation platform, and a visualization terminal; The data acquisition module includes various sensing units installed on the drive mechanism of the amusement ride to collect real-time operating parameters such as torque, current, temperature and vibration. The intelligent assessment platform is used to extract features and model the collected operating parameters, and call artificial intelligence algorithms to generate safety level assessment results corresponding to the driving state; The visualization terminal is used to display the security level assessment results and issue early warning signals in abnormal situations; The aforementioned artificial intelligence algorithms include a state recognition model based on deep learning and a risk prediction model based on probabilistic inference. The state recognition model is used to determine the operating mode of the drive mechanism, and the risk prediction model is used to predict the probability of potential failures.

[0008] The data acquisition module includes various sensing units such as current sensors, torque sensors, temperature sensors, and vibration sensors. The data collected by these sensing units is initially integrated through local acquisition nodes and transmitted to the edge computing unit using industrial-grade communication protocols, such as the CAN bus.

[0009] The edge computing unit is set up on-site at the amusement facility. The edge computing unit preprocesses and extracts features from the raw sensor data. The preprocessing process includes signal filtering, noise suppression, normalization, and outlier removal. Then, the preprocessed data is timestamped to obtain primary data, which is then transmitted to the intelligent evaluation platform in a unified format.

[0010] Furthermore, the intelligent assessment platform includes an algorithm architecture consisting of four layers: a data feature extraction layer, a state recognition layer, a risk prediction layer, and a result fusion and decision layer. The data feature extraction layer is used for deep feature representation. The state recognition layer is used to extract state transition features and, combined with spatial features extracted by CNN, to classify the operating state. After obtaining the state classification results, the risk prediction layer further calls a combined architecture based on Bayesian networks and time series prediction models to assess future operating risks. The result fusion and decision layer weightedly fuses the state recognition results and risk prediction results to generate the final safety level assessment.

[0011] Furthermore, the state recognition layer adopts a state recognition model based on deep learning. The high-dimensional features output by the data feature extraction layer are processed by a convolutional neural network to extract local spatial features. Then, the local spatial features are fed into a long short-term memory network for time-series modeling to obtain the trend of feature changes over time. The state recognition layer is divided into normal operation, mild anomaly, severe anomaly and unknown state.

[0012] Furthermore, the risk prediction layer includes a Bayesian network and a time series prediction model. First, a causal relationship graph between variables is constructed using historical operational data and expert experience. Conditional probabilities are obtained through a Bayesian network, enabling the model to remain adaptive in dynamic environments. The time series prediction model includes a gated recurrent unit and a sequence-to-sequence model. Historical data is extracted using a sliding time window and input into the gated recurrent unit and the sequence-to-sequence model. The time series prediction model outputs predicted values ​​for several future time points through a combination of encoder and decoder.

[0013] Furthermore, the gated loop unit is used to detect long-term dependencies in the sensing signal; the sequence-to-sequence model encodes the historical input sequence into a hidden vector through an encoder and decoder mechanism, and then predicts the signal trend at several future time steps.

[0014] The result fusion and decision-making layer sets pre-weights, and the pre-weights are normalized to obtain the parameter warning weights; the predicted value of the risk prediction layer is multiplied by the corresponding parameter warning weights and then summed to obtain the risk assessment value; the threshold of the risk assessment value is set, and the output result of the state recognition layer is fused with the risk assessment value result to divide it into three safety levels: safety level, warning level, and danger level.

[0015] The visualization terminal receives data from the intelligent evaluation platform and displays it intuitively through a visual interface; the visualization terminal is used to display operational trend analysis and historical data comparison.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention achieves data collection and analysis of the operating status of the drive device by deploying multiple types of sensors and edge computing units; unlike traditional inspection, this system can not only identify the current status in real time, but also dynamically predict future trends through deep learning models, and can predict risks in advance; reducing manual intervention and improving the reliability of safety assessment of amusement facility drives. 2. This invention introduces a prediction model that combines deep learning and probabilistic reasoning, organically combining time series prediction models with Bayesian network causal inference to form a dual mechanism of "trend prediction + causal analysis". This mechanism can accurately predict possible abnormal fluctuations in the future and explain the chain of causes of risk formation. 3. This invention stores all raw material data, proportioning schemes, and performance feedback results in a timestamp structure. Combined with built-in data preprocessing and anomaly correction mechanisms, it supports batch backtracking, process review, and operational trend modeling. This provides production managers with scientific data-driven decision-making support, facilitating the early identification of potential risks such as quality fluctuations and process deviations. Attached Figure Description

[0017] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings; Figure 1 This is a system block diagram of an embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1: As Figure 1 As shown, the present invention provides the following technical solution: It includes a data acquisition module, an intelligent evaluation platform, and a visualization terminal; The data acquisition module includes various sensing units installed on the drive mechanism of the amusement ride to collect real-time operating parameters such as torque, current, temperature and vibration. The data acquisition module includes various sensing units such as current sensors, torque sensors, temperature sensors, and vibration sensors. The data collected by these sensing units is initially integrated through local acquisition nodes and transmitted to the edge computing unit using industrial-grade communication protocols, such as the CAN bus.

[0020] Current sensors are used to collect the current of the drive motor under different operating conditions in order to reflect the motor load; torque sensors are used to obtain changes in drive torque in order to identify overload and other conditions; temperature sensors are used to monitor the temperature rise of major components in order to identify potential overheating risks; and vibration sensors are used to monitor the mechanical vibration generated during operation in order to identify whether there is abnormal vibration or imbalance.

[0021] The intelligent assessment platform is used to extract features and model the collected operating parameters, and call artificial intelligence algorithms to generate safety level assessment results corresponding to the driving state; The visualization terminal is used to display the security level assessment results and issue early warning signals in abnormal situations; The artificial intelligence algorithms include a state recognition model based on deep learning and a risk prediction model based on probabilistic inference. The state recognition model is used to determine the operating mode of the drive mechanism, and the risk prediction model is used to predict the probability of potential failures.

[0022] The edge computing unit is set up on-site at the amusement facility. The edge computing unit preprocesses and extracts features from the raw sensor data. The preprocessing process includes signal filtering, noise suppression, normalization, and outlier removal. Then, the preprocessed data is timestamped to obtain primary data, which is then transmitted to the intelligent evaluation platform in a unified format.

[0023] After the raw sensor data is filtered and noise suppressed, it is normalized and outliers are removed. Then, feature extraction is performed on the normalized data, including: Vibration signals are converted into spectral features using Fast Fourier Transform (FFT) to identify the presence of vibration anomalies; current and torque signals are analyzed using wavelet analysis to extract transient impact features, reflecting sudden changes; temperature data is extracted using sliding window mean and rate of change to obtain trend features; these features are combined into a unified feature vector, with timestamps and sensor identifiers added, and encapsulated into JSON format data. This unified format data is then transmitted to the intelligent assessment platform using an industrial protocol (OPC-UA / Modbus) and / or an IoT protocol.

[0024] Furthermore, the intelligent assessment platform includes an algorithm architecture consisting of four layers: a data feature extraction layer, a state recognition layer, a risk prediction layer, and a result fusion and decision layer. The data feature extraction layer is used for deep feature representation. The state recognition layer is used to extract state transition features and, combined with spatial features extracted by CNN, to classify the operating state. After obtaining the state classification results, the risk prediction layer further calls a combined architecture based on Bayesian networks and time series prediction models to assess future operating risks. The result fusion and decision layer weightedly fuses the state recognition results and risk prediction results to generate the final safety level assessment.

[0025] The data feature extraction layer performs feature extraction on multimodal data, specifically including: For vibration signals, the spectral features obtained from the primary data are first input into a one-dimensional convolutional neural network for feature learning to extract high-frequency components, meshing frequency, and harmonic features. For current and torque signals, wavelet analysis is first performed on primary data to obtain transient impact features, etc., and then the data is input into a two-dimensional convolutional neural network for feature learning to capture pattern changes in local areas of the data, such as current spikes or torque fluctuations, thereby identifying load anomalies and drive imbalances. For temperature data, the primary data reflects the temperature change trend, and then it is input into a one-dimensional convolutional neural network for feature learning to obtain abnormal data features.

[0026] Furthermore, the state recognition layer adopts a state recognition model based on deep learning. The high-dimensional features output by the data feature extraction layer are processed by a convolutional neural network to extract local spatial features. Then, the local spatial features are fed into a long short-term memory network for time-series modeling to obtain the trend of feature changes over time. The state recognition layer is divided into normal operation, mild anomaly, severe anomaly and unknown state.

[0027] The task of the state recognition layer is to transform the high-dimensional feature vectors generated by the data feature extraction layer into classification results of the driving system's operating state; The system employs an LSTM model combined with a convolutional neural network (CNN) at the input end. First, the CNN extracts local spatial patterns, such as changes in vibration frequency peaks or spike features in the current matrix. Then, the local spatial features are fed into the LSTM network for temporal modeling, thereby achieving joint recognition of "spatial features + time dependence". Through this design, the system can simultaneously understand "feature patterns at a certain moment" and "long-term trends", significantly improving classification accuracy.

[0028] Mild anomalies include slight temperature exceedances and short-term increases in vibration amplitude; severe anomalies include high-risk events such as persistently excessive current, unstable torque, and abnormal meshing peaks in the spectrum; when the model cannot provide a high-confidence judgment, the result is classified as "unknown state".

[0029] Furthermore, the risk prediction layer includes a Bayesian network and a time series prediction model. First, a causal relationship graph between variables is constructed using historical operational data and expert experience. Conditional probabilities are obtained through a Bayesian network, enabling the model to remain adaptive in dynamic environments. The time series prediction model includes a gated recurrent unit and a sequence-to-sequence model. Historical data is extracted using a sliding time window and input into the gated recurrent unit and the sequence-to-sequence model. The time series prediction model outputs predicted values ​​for several future time points through a combination of encoder and decoder.

[0030] The risk prediction layer employs a combination of Bayesian networks and time series forecasting models, forming a dual mechanism of causal inference and trend prediction. Bayesian networks are used to establish causal relationships between multidimensional parameters, such as "increased ambient temperature → increased motor winding temperature → current fluctuations → unstable torque → increased vibration." By inferring conditional probability distributions, the system can quantitatively describe the probability of different risk paths occurring, thus providing interpretive support for prediction.

[0031] In time series prediction models, gated recurrent units are used to identify long-term dependencies in sensor signals; sequence-to-sequence models use encoders and decoders to encode historical input sequences into hidden vectors and then predict signal trends at future time steps. The gated loop unit can efficiently capture long-term dependencies in sensor signals. The sequence-to-sequence model structure encodes historical input sequences into hidden vectors through an encoder-decoder mechanism, and then predicts the signal trend at several future time steps. For example, in torque signals, if the system detects that periodic fluctuations are gradually increasing, the sequence-to-sequence model can predict the changes in the amplitude of fluctuations in several future periods and infer potential mechanical fatigue risks accordingly.

[0032] The risk prediction layer uses a Bayesian network to capture causal links between parameters for causal inference, with the aim of answering why problems occur. Then, time series forecasting models are used to predict future trends, which is used for trend prediction to answer the probability of when a problem will occur. Interpretable and predictable risk assessment results are obtained through Bayesian networks and time series prediction models. The risk prediction layer not only improves the system's foresight but also provides a basis for subsequent result integration and decision-making. For example, the probability of mild anomalies within 10 minutes is 35%, the probability of severe anomalies is 12%, and the probability of maintaining a normal state is 53%.

[0033] Furthermore, the results are fused and the decision-making layer sets pre-weights, and the pre-weights are normalized to obtain the parameter warning weights; the predicted value of the risk prediction layer is multiplied by the corresponding parameter warning weights and summed to obtain the risk assessment value; the threshold of the risk assessment value is set, and the output results of the state recognition layer are fused with the risk assessment value results to divide them into three safety levels: safety level, warning level, and danger level.

[0034] The status recognition layer outputs the classification label and confidence level of the current operating status, such as "normal operation, confidence level 92%"; the risk prediction layer outputs the probability of occurrence of various risk events in the future time period, such as "the probability of serious anomalies in the next 10 minutes is 18%". In order to avoid the bias of a single model leading to the distortion of the final judgment, the results are fused with the decision layer in a weighted manner. Since amusement equipment generally undergoes rigorous calibration and testing, its performance is relatively stable and fixed. Therefore, weights are pre-set for different amusement equipment. The weight setting method combines historical data contribution and expert experience. For example, by comparing the contribution of historical data, vibration data with more anomalies is assigned a larger weight, and other data are evaluated in turn. Then, the weight ratio is adjusted based on expert experience. For instance, if expert experience suggests that the amusement equipment often operates at high speeds, vibration data is more important, so the vibration weight is set to twice that of current. Current data is more important than temperature data, so the current weight is set to 1.5 times that of temperature. Finally, the set weights are normalized to obtain the parameter warning weights. The risk assessment value is the sum of the predicted value from the risk prediction layer and the corresponding parameter warning weights. For example, if the warning weight for vibration is 0.50, the warning weight for current is 0.33, and the warning weight for temperature is 0.17, then the risk assessment value R = 0.50*Rv + 0.33*Rc + 0.17*Rt, where Rv, Rc, and Rt are the output risk values ​​for vibration, current, and temperature, respectively, obtained through the risk prediction layer.

[0035] The thresholds for risk assessment values ​​are set based on expert experience. Among them, the safety level is: all monitoring indicators are within the normal range and the probability of risk prediction is lower than the set threshold; the alert level is: the current state may have a slight abnormality, or the probability of risk prediction will increase significantly in the future but has not yet exceeded the danger threshold; the danger level is: there is a serious abnormality at present, or the probability of failure in the future prediction exceeds the high-risk threshold, and immediate measures must be taken.

[0036] For example, in a large roller coaster project, the drive motor and gearbox are critical components during operation. The system is equipped with vibration sensors, current sensors, and temperature sensors on these components. During testing, the system detected slight fluctuations in the motor current during acceleration. The status identification layer classifies this as a "minor anomaly" with a confidence level of 78%. Subsequently, the risk prediction layer further analyzes the data, predicting a 42% probability of continued current fluctuations and a 25% risk of torque fluctuations within the next 5 minutes. Based on this, the result fusion and decision-making layer outputs an "alert level," prompting engineers to check the motor load and lubrication status. During actual maintenance, engineers discovered dust accumulation on the motor cooling fan, leading to poor heat dissipation. Timely cleaning prevented the risk of motor overheating and shutdown.

[0037] To ensure the accuracy of result integration and decision-making, continuous adjustments and updates are necessary. For example, when operations and maintenance personnel confirm or correct the system output, this feedback will be recorded and used to optimize weight allocation and threshold settings. If a certain "alert level" is ultimately confirmed to be normal, the model's weights in similar situations should be adjusted appropriately to reduce the false alarm rate.

[0038] Based on the above, the visualization terminal receives data from the intelligent evaluation platform and displays it intuitively through a visualization interface; the visualization terminal is used to display operational trend analysis and historical data comparison.

[0039] This invention achieves data collection and analysis of the operating status of the drive device by deploying multiple types of sensors and edge computing units. Unlike traditional inspections, this system can not only identify the current status in real time, but also dynamically predict future trends through deep learning models, enabling early prediction of risks. This reduces human intervention and improves the reliability of safety assessments for amusement park rides. This invention introduces a prediction model that combines deep learning and probabilistic reasoning, organically combining time series prediction models with Bayesian network causal inference to form a dual mechanism of "trend prediction + causal analysis". This mechanism can accurately predict possible abnormal fluctuations in the future and explain the chain of causes of risk formation. This invention stores all raw material data, proportioning schemes, and performance feedback results in a timestamp structure. Combined with built-in data preprocessing and anomaly correction mechanisms, it supports batch backtracking, process review, and operational trend modeling. This provides production managers with scientific data-driven decision-making support, facilitating the early identification of potential risks such as quality fluctuations and process deviations. The preferred embodiments of this invention disclosed above are merely illustrative. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to specific implementation methods. Obviously, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, enabling those skilled in the art to better understand and utilize it. This invention is limited only by the claims and their full scope and equivalents.

Claims

1. An artificial intelligence-based intelligent safety assessment system for amusement rides, characterized in that, The system includes a data acquisition module, an intelligent assessment platform, and a visualization terminal. The data acquisition module comprises multiple sensor units installed on the amusement ride's drive mechanism to collect real-time operating parameters such as torque, current, temperature, and vibration. The intelligent assessment platform extracts features and models the collected operating parameters, and uses artificial intelligence algorithms to generate safety level assessment results corresponding to the drive state. The visualization terminal displays the safety level assessment results and issues warning signals in abnormal situations. The artificial intelligence algorithms include a deep learning-based state recognition model and a probability inference-based risk prediction model. The state recognition model determines the drive mechanism's operating mode, and the risk prediction model predicts the probability of potential failures.

2. The intelligent safety assessment system for amusement rides based on artificial intelligence according to claim 1, characterized in that, The data acquisition module includes various sensing units such as current sensors, torque sensors, temperature sensors, and vibration sensors. The data collected by these sensing units is initially integrated through local acquisition nodes and then transmitted to the edge computing unit using an industrial-grade communication protocol.

3. The artificial intelligence-based intelligent assessment system for the safety of amusement rides according to claim 2, characterized in that, The edge computing unit is set up at the amusement facility site, and the edge computing unit preprocesses and extracts features from the raw sensor data; The preprocessing process includes signal filtering, noise suppression, normalization, and outlier removal. Then, the preprocessed data is timestamped to obtain primary data, which is then transmitted to the intelligent evaluation platform in a unified format.

4. The intelligent safety assessment system for amusement rides based on artificial intelligence as described in claim 1, characterized in that, The intelligent assessment platform includes an algorithm architecture consisting of four layers: a data feature extraction layer, a state recognition layer, a risk prediction layer, and a result fusion and decision-making layer. The data feature extraction layer is used for deep feature representation; The state recognition layer is used to extract state transition features and combine them with spatial features extracted by CNN to classify the operating state; after obtaining the state classification results, the risk prediction layer further calls the combined architecture based on Bayesian network and time series prediction model to assess future operating risks. The results fusion and decision-making layer weightedly fuses the state identification results with the risk prediction results to generate the final security level assessment.

5. The intelligent safety assessment system for amusement rides based on artificial intelligence according to claim 4, characterized in that, The state recognition layer adopts a deep learning-based state recognition model. The high-dimensional features output by the data feature extraction layer are processed by a convolutional neural network to extract local spatial features. Then, the local spatial features are fed into a long short-term memory network for time-series modeling to obtain the trend of feature changes over time. The state recognition layer is divided into normal operation, mild anomaly, severe anomaly and unknown state.

6. The intelligent safety assessment system for amusement rides based on artificial intelligence according to claim 4, characterized in that, The risk prediction layer comprises a Bayesian network and a time series prediction model. First, a causal relationship graph between variables is constructed using historical operational data and expert experience. Conditional probabilities are obtained through a Bayesian network, enabling the model to remain adaptive in dynamic environments. The time series prediction model includes a gated recurrent unit and a sequence-to-sequence model. Historical data is extracted using a sliding time window and input into the gated recurrent unit and the sequence-to-sequence model. The time series prediction model outputs predicted values ​​for several future time points through a combination of encoder and decoder.

7. The intelligent safety assessment system for amusement rides based on artificial intelligence according to claim 6, characterized in that, The gated loop unit is used to detect long-term dependencies in the sensing signal; the sequence-to-sequence model encodes the historical input sequence into a hidden vector through an encoder and decoder mechanism, and then predicts the signal trend at several future time steps.

8. The intelligent safety assessment system for amusement rides based on artificial intelligence according to claim 6, characterized in that, The results are integrated with the pre-set weights of the decision-making layer, and the pre-set weights are normalized to obtain the parameter warning weights; the predicted values ​​of the risk prediction layer are multiplied by the corresponding parameter warning weights and then summed to obtain the risk assessment value; A threshold for the risk assessment value is set, and the output results of the status recognition layer are merged with the risk assessment value results to divide them into three safety levels: safety level, alert level, and danger level.

9. The intelligent safety assessment system for amusement rides based on artificial intelligence according to claim 1, characterized in that, The visualization terminal receives data from the intelligent evaluation platform and displays it intuitively through a visual interface; the visualization terminal is used to display operational trend analysis and historical data comparison.

Images