A deep learning-based LED health monitoring system and method

Abstract

A kind of LED health monitoring system and method based on deep learning, including central control module, LED drive module, LED lighting module, state monitoring module, weather sensing module, man-machine interaction module and deep learning module, display warning module;LED drive module, weather sensing module, man-machine interaction module, display warning module and deep learning module are all connected to central control module;LED drive module connects LED lighting module, and state monitoring module is arranged between LED lighting module and central control module;The multiple characteristic parameters of the LED lighting system are monitored in the application, the health status of LED light source and power supply can be reflected simultaneously.The electrical characteristics of LED under different room temperature and current are established by using the deep learning method in the application, the electrical characteristics of LED under different working conditions can be accurately reflected, and reference is provided for health status monitoring.

Description

Technical Field

[0001] This invention relates to the field of intelligent lighting technology, and in particular to an LED health monitoring system and method based on deep learning. Background Technology

[0002] LEDs have gradually become the mainstream product in the lighting industry due to their numerous advantages such as high brightness, energy saving, and low price. Although the reliability of LED devices has gradually improved with continuous advancements in technology and manufacturing processes, LEDs still have many unique reliability issues compared to other lighting fixtures.

[0003] (1) LEDs are point-source light sources, and the heat generated is concentrated in a small area. They have a high power density, are difficult to dissipate, and are prone to increasing junction temperature and device degradation.

[0004] (2) After a period of use, the light color performance of LEDs decreases as the devices age;

[0005] (3) The structure of LED driver power supply is more complex than other types of power supply structure, and the power supply failure and failure caused by component aging are more serious.

[0006] Therefore, although the theoretical lifespan of LEDs can reach 100,000 hours, in practical applications, temperature, humidity, and current are all important factors affecting their reliability and lifespan, causing their lifespan to be significantly shorter than expected. Thus, research on the reliability of LED lighting systems and monitoring their health status are particularly important.

[0007] For high-power LEDs, especially in important public places such as stages and stadiums, it is of great significance to detect and evaluate the health status of LEDs.

[0008] Traditional LED health monitoring methods often rely on empirical or statistical models combined with test data to assess the health status of LEDs. The biggest problem with this method is that it cannot account for the differences between individual LEDs, nor can it incorporate the historical degradation trends of individual LEDs to more accurately assess their health status. Summary of the Invention

[0009] The purpose of this invention is to provide an LED health monitoring system and method based on deep learning to solve the above-mentioned problems.

[0010] To achieve the above objectives, the present invention adopts the following technical solution:

[0011] A deep learning-based LED health monitoring system includes a central control module, an LED driver module, an LED lighting module, a status monitoring module, a weather sensor module, a human-computer interaction module, a display and early warning module, and a deep learning module. The LED driver module, weather sensor module, human-computer interaction module, display and early warning module, and deep learning module are all connected to the central control module. The LED driver module is connected to the LED lighting module, and a status monitoring module is provided between the LED lighting module and the central control module.

[0012] The weather sensor module is used to collect real-time weather data, including outdoor brightness, temperature, and humidity.

[0013] The human-computer interaction module is used to input LED lighting requirements, including lighting time period and brightness, the alarm threshold of the early warning module, and the sampling interval of the status detection module;

[0014] The deep learning module is used to store deep learning models and calculate predicted values ​​for LED lighting modules based on weather data and user input.

[0015] The status monitoring module is used to monitor the real-time operating status data of the LED lighting module;

[0016] The central control module is used to output control signals to the LED lighting module based on data from the human-machine interaction module or weather sensor module, and to process data transmitted to it from the modules connected to it.

[0017] The LED driver module is used to output signals that control the brightness of the LED.

[0018] The display and early warning module is used to display weather parameters, predicted values ​​of LED lighting modules, and real-time operating status data;

[0019] The LED lighting module is used to emit light of corresponding brightness according to the control signal from the LED driver module.

[0020] Furthermore, if the error between the detected value and the predicted value of the LED lighting module displayed by the warning module exceeds the alarm threshold of the warning module, an alarm signal will be issued.

[0021] Furthermore, the alarm threshold of the display warning module is set at 3% to 15% of the deviation between the detected value and the predicted value of the LED lighting module.

[0022] Furthermore, the predicted values ​​include the current, voltage drop, and case temperature of the LED lighting module; the operating status data includes the current, voltage drop, and case temperature of the LED lighting module.

[0023] Furthermore, the sampling interval of the status detection module is 5 min to 30 min.

[0024] Furthermore, a deep learning-based LED health monitoring method includes the following steps:

[0025] Store the deep learning model in the deep learning module;

[0026] The human-computer interaction module inputs the LED lighting requirements, displays the alarm threshold of the warning module, and the test time interval of the status detection module.

[0027] The weather sensor module collects real-time weather data at user-input intervals; all data is then aggregated into the central control module.

[0028] The central control module calculates the drive current value of the LED lighting module based on the aggregated data from the human-machine interaction module and the weather sensor module, and outputs the control signal of the LED lighting module.

[0029] The status monitoring module collects the real-time working status of the LED lighting module periodically according to the detection time interval input by the human-computer interaction module;

[0030] The deep learning module calculates the predicted value of the LED lighting module based on real-time weather parameters and the input value of the human-computer interaction module.

[0031] The central control module compares the test values ​​and predicted values ​​of the LED lighting module. If the difference between the two values ​​is less than a threshold, the unwarranted test data is added to the historical data of the deep learning module to update the deep learning model; if the difference between the two values ​​is greater than the threshold, an alarm is issued.

[0032] The display and warning module shows weather parameters, predicted and detected values ​​from the LED lighting module, offset, and alarms.

[0033] Furthermore, deep learning models are built using the following steps:

[0034] Step 1: When the LED lighting module leaves the factory, collect the voltage drop and brightness under different room temperature, humidity and operating current, and reconstruct this data into time series data for use as historical data for model training.

[0035] Step 2: Using room temperature, humidity, outdoor brightness, and LED module luminance as inputs, and using the LED module current, voltage drop, and casing temperature as outputs, construct a long short-term memory deep learning network.

[0036] Step 3: Using the historical data as training data, train a deep learning model using a deep learning framework;

[0037] Step 4: Calculate the predicted values ​​of the trained model for the training input data, calculate the error between the predicted values ​​and the measured values, and use a genetic algorithm to find the minimum value of the error between the predicted values ​​and the measured values ​​in order to obtain the optimal values ​​of the network parameters.

[0038] Step 5: The deep learning module calculates the predicted values ​​of the current, voltage drop, and casing temperature of the LED lighting module based on the real-time weather parameters and the input values ​​of the human-computer interaction module.

[0039] Step 6: The status monitoring module collects the current, voltage drop, and brightness of the LED lighting module and compares them with the predicted values. If the error does not exceed the warning threshold, it is added to the historical data; if the error exceeds the threshold, the display warning module outputs a warning signal.

[0040] Step 7: Use the test data obtained in Step 6 to train and update the deep learning model.

[0041] Furthermore, the deep learning framework is the LSTM regression model, with 128 to 512 neurons, a learning rate of 0.001 to 0.05, and 200 to 2000 iterations.

[0042] Compared with the prior art, the present invention has the following technical effects:

[0043] This invention monitors multiple characteristic parameters of LED lighting systems, simultaneously reflecting the health status of both the LED light source and the power supply. Employing deep learning methods and based on historical test data, this invention establishes a system that accurately reflects the electrical characteristics of LEDs under different operating conditions, providing a reference for health status monitoring. This allows for full consideration of the individual differences of each LED, providing users with more precise information on LED health status and remaining lifespan.

[0044] This invention uses deep learning to establish the electrical characteristics of LEDs under different room temperature currents, which can accurately reflect the electrical characteristics of LEDs under different operating conditions and provide a reference for health status monitoring.

[0045] This invention uses new test data to update deep learning networks, making the model more adaptable and portable. Attached Figure Description

[0046] Figure 1 This is a system structure diagram of the present invention.

[0047] Figure 2 This is a flowchart of the present invention. Detailed Implementation

[0048] The present invention will be further described below with reference to the accompanying drawings:

[0049] A deep learning-based LED health monitoring system includes a central control module, an LED driver module, an LED lighting module, a status monitoring module, a weather sensing module, a display and early warning module, a human-computer interaction module, and a deep learning module.

[0050] The central control module is interconnected with the weather sensing module, display and early warning module, human-computer interaction module, LED driver module, deep learning module, and status detection module.

[0051] The LED lighting module is interconnected with the LED driver module and the status detection module.

[0052] A deep learning-based method for LED health monitoring includes the following steps:

[0053] Step 1: The user inputs their brightness adjustment needs, daily lighting time period, and test interval of the status detection module through the human-computer interaction module.

[0054] Step 2: The weather sensor module collects real-time weather data every 5 to 15 minutes, including outdoor brightness, temperature, humidity, etc.; all data is then aggregated into the central control module.

[0055] Step 3: Based on the summarized data from Step 1, the central control module obtains the drive current value of the LED lighting module and outputs the control signal for the LED lighting module.

[0056] Step 4: The status monitoring module collects the real-time operating status of the LED lighting module at regular intervals according to the test time intervals described in Step 1, including the current, voltage drop, and casing temperature of the LED lighting module, and adds the test data to the historical data.

[0057] Step 5: The central control module uses deep learning algorithms to calculate the predicted values ​​of current, voltage drop, and casing temperature of the LED lighting module based on real-time weather parameters.

[0058] Step 6: Calculate the offset between the test values ​​and predicted values ​​of the LED lighting module's current, voltage drop, and casing temperature;

[0059] Step 7: Output the offset values ​​of the LED lighting module current and voltage drop to the display and warning module.

[0060] The deep learning model is established using the following steps:

[0061] Step 1: Collect the voltage drop and brightness of the LED lighting module at different temperatures and current values ​​before use, and use this data as historical data;

[0062] Step 2: Construct a deep learning framework;

[0063] Step 3: Using the historical data as training data, train the deep learning model using a deep learning framework;

[0064] Step 4: Once the accuracy requirements are met, training is complete;

[0065] Step 5: Collect the current, voltage drop, and brightness of the LED lighting module during use and add them to the historical data;

[0066] Step 6: Use the test data obtained in Step 5 to train and update the deep learning model.

[0067] The deep learning framework used is an LSTM regression model. The inputs are air temperature and current values, and the outputs are voltage drop, brightness, and shell temperature. The hidden layer has 4*(m*n+n) neurons, where m and n are the input and output functions, respectively. The loss function is set to mean absolute error, and the optimization function is set to the admin function.

[0068] Example 1:

[0069] Store the deep learning model in the deep learning module;

[0070] The human-computer interaction module inputs the LED lighting requirements, displays the alarm threshold of 5% in the warning module, and sets the test interval of 30 minutes in the status detection module.

[0071] The weather sensor module collects real-time weather data at user-input intervals; all data is then aggregated into the central control module.

[0072] The central control module calculates the drive current value of the LED lighting module based on the aggregated data from the human-machine interaction module and the weather sensor module, and outputs the control signal of the LED lighting module.

[0073] The status monitoring module collects the real-time working status of the LED lighting module periodically according to the detection time interval input by the human-computer interaction module;

[0074] The deep learning module calculates the predicted value of the LED lighting module based on real-time weather parameters and the input value of the human-computer interaction module.

[0075] The central control module compares the test values ​​and predicted values ​​of the LED lighting module. If the difference between the two values ​​is less than a threshold, the unwarranted test data is added to the historical data of the deep learning module to update the deep learning model; if the difference between the two values ​​is greater than the threshold, an alarm is issued.

[0076] The display and warning module shows weather parameters, predicted and detected values ​​from the LED lighting module, offset, and alarms.

[0077] The number of neurons, learning rate, and number of iterations in the deep learning algorithm are set to 128, 0.0005, and 200, respectively.

[0078] This embodiment uses a longer sampling interval, fewer neurons and fewer iterations, resulting in lower network complexity, faster computation, and fewer resource requirements.

[0079] Example 2:

[0080] Store the deep learning model in the deep learning module;

[0081] The human-computer interaction module inputs the LED lighting requirements, displays the alarm threshold of 10% in the warning module, and sets the test interval of 20 minutes in the status detection module.

[0082] The number of neurons, learning rate, and number of iterations in the deep learning algorithm are set to 256, 0.001, and 500, respectively.

[0083] This embodiment uses compromise parameters, which can balance computing power requirements and model accuracy.

[0084] Example 3:

[0085] Store the deep learning model in the deep learning module;

[0086] The human-computer interaction module inputs the LED lighting requirements, displays the alarm threshold of 15% in the warning module, and sets the test interval of 10 minutes in the status detection module.

[0087] The number of deep learning neurons, learning rate, and number of iterations are set to 512, 0.01, and 1000, respectively.

[0088] This embodiment uses a smaller sampling interval to obtain more test data, and uses a larger number of neurons and iterations to make the model larger, which can fit more complex LED degradation health characteristics and obtain more accurate prediction results, but at the same time requires more computing resources.

Claims

1. A deep learning based LED health monitoring system, characterized in that, The system comprises a central control module, an LED driving module, an LED lighting module, a state monitoring module, a weather sensing module, a human-computer interaction module, a display warning module and a deep learning module; the LED driving module, the weather sensing module, the human-computer interaction module, the display warning module and the deep learning module are all connected to the central control module; the LED driving module is connected to the LED lighting module, and the state monitoring module is arranged between the LED lighting module and the central control module; The weather sensing module is used to collect real-time weather data including outdoor brightness, air temperature and humidity; The human-computer interaction module is used to input LED lighting requirements including lighting time period and brightness, alarm threshold of the warning module and sampling interval of the state detection module; The deep learning module is used to store a deep learning model and calculate predicted values of the LED lighting module according to weather data and user input values; The state monitoring module is used to monitor real-time working state data of the LED lighting module; The central control module is used to output control signals of the LED lighting module according to data of the human-computer interaction module or the weather sensing module, and process data transmitted by the modules connected to it; The LED driving module is used to output output signals for controlling light emitting brightness of the LED; The display warning module is used to display weather data, predicted values of the LED lighting module and real-time working state data; The LED lighting module is used to emit light with corresponding brightness according to control signals of the LED driving module; The deep learning model is established by the following steps: Step 1: when the LED lighting module is shipped, collect pressure drop and brightness under different room temperature, humidity and working current, reconstruct the data into time series data, and use the data as historical data for model training; Step 2: construct a long short-term memory deep learning network with room temperature, humidity, outdoor brightness and LED module light emitting brightness as input, and LED module current, pressure drop and shell temperature as output; Step 3: use the deep learning framework to train the deep learning model with the historical data as training data; Step 4: calculate predicted values of the model obtained by training for training input data, calculate errors between the predicted values and measured values, use a genetic algorithm to find the minimum value of the errors between the predicted values and the measured values, and obtain optimal values of network parameters; Step 5: the deep learning module calculates predicted values of the LED lighting module's current, pressure drop and shell temperature according to real-time weather data and input values of the human-computer interaction module; Step 6: the state monitoring module collects the LED lighting module's current, pressure drop and shell temperature, compares them with the predicted values, and if the errors do not exceed the warning threshold, supplements the historical data; If the errors exceed the threshold, the display warning module outputs a warning signal; Step 7: use the test data obtained in step 6 to train and update the deep learning model; The deep learning module calculates predicted values of the LED lighting module according to real-time weather data and input values of the human-computer interaction module; The predicted values are predicted values of the LED lighting module's current, pressure drop and shell temperature.

2. The deep learning based LED health monitoring system of claim 1, wherein, If the error between the detected value and the predicted value of the LED lighting module displayed by the display warning module exceeds the alarm threshold of the display warning module, an alarm signal is sent out.

3. The deep learning based LED health monitoring system of claim 2, wherein, The alarm threshold of the display warning module is 3% to 15% of the error between the detected value and the predicted value of the LED lighting module.

4. The deep learning based LED health monitoring system of claim 1, wherein, The predicted value includes the current, voltage drop and shell temperature of the LED lighting module; and the working state data includes the current, voltage drop and shell temperature of the LED lighting module.

5. The deep learning based LED health monitoring system of claim 4, wherein, The sampling interval of the state detection module is 5 minutes to 30 minutes. 6.A method for LED health monitoring based on deep learning, characterized in that, The deep learning-based LED health monitoring system according to any one of claims 1 to 5 comprises the following steps: The deep learning model is stored in the deep learning module; The lighting requirements of the LED, the alarm threshold of the display warning module and the test time interval of the state detection module are input through the human-computer interaction module; The weather sensing module collects real-time weather data every time interval input by the user; and the real-time weather data is summarized to the central control module; The central control module calculates the driving current value of the LED lighting module according to the summarized data of the human-computer interaction module and the weather sensing module, and outputs the control signal of the LED lighting module; The state detection module collects the real-time working state of the LED lighting module at regular intervals according to the test time interval input by the human-computer interaction module; The deep learning module calculates the predicted value of the LED lighting module according to the real-time weather data and the input value of the human-computer interaction module; The central control module compares the measured value and the predicted value of the LED lighting module, and if the difference between the two values is less than the threshold value, the non-warning measured value is supplemented to the historical data of the deep learning module to update the deep learning model; and if the difference between the two values is greater than the threshold value, an alarm is sent out; The display warning module displays the weather data, the predicted value and the measured value of the LED lighting module, the difference between the measured value and the predicted value, and the alarm.

7. The LED health monitoring method based on deep learning according to claim 6, characterized in that, The deep learning framework is an LSTM regression model, the number of neurons is 128 to 512, the learning rate is 0.001 to 0.05, and the iteration number is 200 to 2000.

Images