Synergistic effect of artificial intelligence and new real-time disassembly sensors: Overcoming limitations and expanding application scope

2.1 Reflection of synergistic effect

The new real-time disassembly sensor is a sensor with a real-time monitoring function and can be easily disassembled and installed. It is designed for rapid installation in equipment or systems to monitor various physical or chemical parameters in real time [17,18]. The collaborative cooperation between artificial intelligence and new real-time disassembly sensors has been reflected in various industries in multiple aspects.

First, a large amount of real-time data can be provided through sensors, and corresponding artificial intelligence technologies can use this data for pattern recognition [19], anomaly detection [20], trend analysis [21], and so on. Second, artificial intelligence technology analyzes the historical data collected by sensors to predict potential equipment failures in advance, so that staff can perform maintenance promptly, and reduce equipment downtime and repair costs, which is difficult to achieve with a single sensor. Third, sensors combined with artificial intelligence technology can analyze and mine device operation data, providing intelligent decision support for decision-makers. Fourth, artificial intelligence technology can visually display device data and facilitate real-time monitoring and control by transmitting sensor data in real-time to the system. In summary, the synergistic effect of artificial intelligence and new real-time disassembly sensors can achieve real-time monitoring, prediction, and optimization of equipment status, improve equipment reliability, safety, and efficiency, and bring significant technological progress and economic benefits to industrial production and equipment maintenance.

2.2 Potential applications of artificial intelligence and new real-time disassembly sensors

The application range of individual sensors is often limited, but after combining artificial intelligence technology, they are gradually applied in various industries. The synergistic effect of artificial intelligence and new real-time disassembly sensors has a wide range of potential applications in various fields, mainly manifested in manufacturing, energy, transportation, agriculture, healthcare, and other industries.

In the manufacturing industry, real-time disassembly of sensors and the synergistic effect of artificial intelligence can achieve real-time monitoring and predictive maintenance of equipment. Sensors collect device operation data, and artificial intelligence algorithms can analyze this data to achieve prediction of device status and fault diagnosis, thereby improving production efficiency and product quality. In the energy industry, real-time disassembly of sensors combined with artificial intelligence technology can be used to monitor the operational status of facilities such as power equipment and oil and gas pipelines [22,23]. In transportation, real-time disassembly of sensors and the synergistic effect of artificial intelligence can be used to monitor the operation status and driving environment of vehicles [24].

For agriculture, real-time disassembly of sensors and artificial intelligence technology can be used to monitor agricultural parameters such as soil moisture and crop growth [25]. In healthcare, real-time disassembly of sensors and the synergistic effect of artificial intelligence can be used to monitor the physiological parameters and health status of patients [26].

3.1 GRU model

GRU is a variant of recurrent neural network (RNN). It solves the problem of exploding and disappearing weight levels in traditional RNN models, as well as the problem of applying too many parameters and making training difficult for long short-term memory (LSTM) models [27,28]. The GRU model combines the forget gate and input gate together, correspondingly reducing the number of gates. This ensures the memory ability of the GRU model while improving the training performance of the network.

The reset gate is mainly used to control the retention of previous memories, and the calculation formula is

Among them, U js corresponds to the weight matrix, c ks corresponds to the bias vector, and y t represents the input at time t.

The calculation formula for updating the door is

Among them, k t − 1 represents the hidden state at time t−1, and γ represents the sigmoid activation function.

The calculation formula for hidden doors is

The calculation of the hidden state at time t is

The calculation formula for the output generator fault type is

3.2 PSO

To optimize the GRU model, the particle swarm optimization algorithm is now applied for optimization [29]. The specific steps are as follows:

1. The parameters of the particle swarm can be initialized and corresponding PSO parameters can be set. In this experiment, the number of particle swarms is set to 100, the number of iterations is 30, and the search dimension is set to 4.

2. The position and velocity of the particle swarm can be initialized, and the particle position can be set using a random initialization method. The representation of the particle position is shown in formula (6).

   (6) Y j = ( w , n , k 1 , k 2 , k 3 ) .

   Among them, w represents the learning rate of the GRU model, ranging from 0.001 to 0.005, and n represents the number of iterations of the GRU model.

   The representation of particle velocity is

   (7) P j = ( P j 1 , P j 2 , P j 3 , P j 4 , P j 5 ) .

3. Set the search time step of the GRU model to 1 in this experiment.

4. The fitness function of PSO can be determined, and the GRU model framework can be constructed using initialized particles. The calculation of its fitness function is

   (8) g ( z ) = 1 A ∑ j = 1 A ( z − z ′ ) 2 .

   Among them, A represents the number of training batches each time; z represents the actual fault situation of the corresponding generator, and z ′ represents the predicted fault situation of the corresponding generator.

5. Continuing iterative calculation: By calculating the position of particles after each iteration and continuously adjusting the speed and position of the example, the optimal result is output. The calculation of speed update is shown in formula (9), and the calculation of position update is

   (9) P j t + 1 = u P j t + d 1 q 1 ( E j t − Y j t ) + d 2 q 2 ( E h t − Y j t ) .

   Among them, u represents inertia weight, d 1 and d 2 both represent learning factors, E j t represents the optimal position of the t-th particle, and E h t represents the global optimal position of the t-th particle.

   (10) Y j t + 1 = Y j t + P j t + 1 Time .

   Among them, Time represents the time factor.

   The time factor calculation formula is

   (11) Time = 1 t O .

6. Finally, the initial parameters were optimized using the PSO algorithm, and the GRU model was trained. After training, the fault types of the generator were classified.

3.3 SVM algorithm

In the GRU model, the output layer uses the traditional Softmax activation function for classification, but can only perform single classification, while the generator fault type in this experiment belongs to multi-classification. To improve the predictive performance of generator fault types, SVM is now applied to optimize the output layer of the GRU model. The calculation formula for SVM is

The penalty term is represented by D, and δ j represents the slack variable.

The radial basis function is a commonly used kernel function, which is mainly used to map data to a high-dimensional space in SVM so that a linearly separable hyperplane can be found in the high-dimensional space for classification.

In this experiment, the radial basis function was chosen as the kernel function, and its calculation formula is

The calculation formula for the classification function corresponding to the SVM algorithm is

where sgn represents the sign function, c represents the bias term, β j represents the Lagrange multiplier, and l ( y i , y j ) represents the kernel function. The sign function sgn is introduced to convert the continuous decision value into a discrete category label, and its output directly determines which category the input sample y belongs to.

4.1 Experimental environment

Hardware equipment: 13th Gen Intel (R) Core(TM) i7-13700HX CPU (central processing unit), 32GB memory, memory speed 4800 MHz, NVIDIA GeForce RTX 4080 Laptop GPU (graphics processing unit).

New real-time disassembly sensors: red thermal external imager FLIR T540, MEMS vibration sensor STMicroelectronics LSM6DS3, nano-insulation sensor, acoustic flow sensor Honeywell SmartLine ST75, radio flow sensor AmpSense AS200, and nano dust sensor.

Software equipment: Microsoft Windows 11 operating system, Python software.

4.2 Experimental data

The data for this experiment comes from the generator equipment of a certain enterprise in Beijing. The collected data includes generator bearing vibration value, bearing temperature, noise value, insulation resistance value, winding temperature, rotor vibration value, rotor temperature, coolant flow rate, coolant temperature, current value, internal carbon powder accumulation degree of the generator, historical fault frequency, oil temperature, environmental temperature, fuel consumption rate, output voltage, output power, etc. The experiment collected data every 10 s, and a total of 2,340 pieces of data were collected through the new real-time disassembly sensor. The data partitioning method is to use the tenfold cross-validation method [30] to partition the generator equipment data. The experiment takes turns conducting experiments and taking the average of the results as the final result. Part of the experimental data is shown in Table 1.

In Table 1, A represents the data set, B represents the bearing vibration value (mm), C represents the bearing temperature (°C), D represents the noise value (dB), E represents the insulation resistance value (Ω), F represents the winding temperature (°C), G represents the rotor vibration value (mm), H represents the rotor temperature (°C), I represents the coolant flow (m³/h), J represents the coolant temperature (°C), K represents the current value (A), and L represents the degree of toner aggregation inside the generator (%).

In Table 1, to better compare the generator fault monitoring results in this experiment, four different fault time data are randomly selected from the original data for display (1–2 is a period, 3–5 is a period, 6–8 is a period, and 9–10 is a period). The normal range of measurement parameters and indicators of the new real-time disassembly sensor is shown in Table 2.

From Table 2, it can be seen that the normal range of effective vibration values for bearings is 0.05–0.2 mm, and the normal range of bearing temperature is 58–80°C. The vibration range of noise generated by the generator under normal conditions is 70–85 dB, and the normal temperature range of the winding is 70–90°C. The normal range of rotor vibration values is 0.03–0.15 mm, and the normal range of rotor temperature is 60–83°C. The coolant temperature ranges from 26 to 43°C, and the normal range of current values is 150–300 A. The degree of carbon powder accumulation inside the generator is usually less than 5%.

4.3 Data preprocessing

4.4 Experimental process

This experiment is first based on the collection of new real-time disassembly sensors installed on the generator. The data is processed for missing values, normalized, and feature engineering. Then, the collected data from the sensors is divided and added to the optimized GRU model for training. Finally, the performance of the model is verified on the test set.

4.5 Experimental results

Evaluating indicator:

Among them, accuracy:

precision:

recall:

f1:

TP represents true positive; TN represents true negative; FP represents false positive; FN represents false negative.

After the above experimental steps, the experimental results are shown in Table 4.

In Table 4, data groups 3–5 show that the actual generator in data group 3 is in a normal state, which is also predicted by this model to be normal. In data group 4, it is actually normal, but the model predicts a carbon brush fault precisely because the model detected a current value of 268.2 A, and the accumulation of carbon powder inside the generator was as high as 4.9%, only 0.1% below the alarm limit. After 10 s, in data group 5, the generator actually experienced a carbon brush fault, and the model predicted it to be a carbon brush fault, with a current value of 340.4 A. The accumulation of carbon powder inside the generator reached 6.8%, which is higher than the normal limit of 1.8%. It can be seen that the new real-time disassembly sensor, with the introduction of artificial intelligence technology, can predict generator faults in advance, repair them promptly, and avoid downtime. In addition, in data groups 6–8, the actual generator in data group 6 is in a normal state. After prediction by this model, it was normal. However, after 10 s, the actual state of the generator was normal, and the model predicted a bearing failure. It is precisely because the model detected that the bearing vibration value reached the minimum limit of 0.05 mm, and the bearing temperature reached 78°C, with a noise value of 84.8 dB. All three values are close to the critical value. After another 10 s, the generator experienced a bearing failure, and the model predicted it to be a bearing failure. The vibration value of the bearings has significantly decreased, only to 0.01 mm. The temperature of the generator bearings reached as high as 91.1°C, and the noise level reached 88.2 dB, seriously exceeding the limit value. Based on the comparison of the above prediction results, it can be concluded that combining artificial intelligence technology with real-time sensors can reduce downtime and enable staff to maintain it promptly.

4.6 Application range expansion results of artificial intelligence and new real-time dismantling sensors

1. To further explore the application scope of the model, this article studies the synergistic effect of artificial intelligence technology and new real-time disassembly sensors. It is now being applied in the energy industry, transportation, agriculture, and healthcare industries. The specific implementation steps are as follows:

   In a certain energy enterprise, a new type of real-time disassembly sensor is deployed to collect real-time vibration, temperature, noise, and other data of generator equipment, and the collected data is transmitted to the data center for storage and analysis. Based on the collected data, the optimized GRU model can be applied to predict the possible types and times of faults that may occur in generator equipment.

2. New real-time disassembly sensors can be installed on cars to monitor parameters such as vibration, temperature, and noise of the vehicle. It can collect real-time vehicle status data and transmit it to the central control center for analysis. Based on the collected data, an optimized GRU can be used to establish an anomaly detection model for vehicle status, which can verify that when abnormal vehicle status is detected, timely warnings are issued, and corresponding measures are taken to reduce the probability of accidents.

3. By deploying a new type of real-time dismantling sensor in rural farmland, parameters such as soil moisture, temperature, and crop growth can be monitored in real time. The optimized GRU model can be used to analyze monitoring data in real time, provide real-time status and prediction information on crop growth, and achieve precise irrigation and fertilization of farmland. Water and nutrient supply can be adjusted according to crop demand to improve crop yield and quality.

4. The medical and health industry chooses a hospital in Nanchang City. New real-time detachable sensors can be used to monitor the physiological parameters of patients, such as heart rate, body temperature, blood pressure, etc., and the monitoring data can be transmitted to the medical system to achieve real-time monitoring of the patient’s health status. Based on patient monitoring data and historical health records, an optimized GRU can be used to establish disease prediction and diagnostic models, timely predict potential disease risks for patients, and provide personalized diagnosis and treatment recommendations.

After the above experiment, the specific effect is shown in Figure 1.

Figure 1

Application range expansion results.

In Figure 1, the application of artificial intelligence and new real-time disassembly sensors in the energy industry, transportation, agriculture, and healthcare industries are shown in sequence. From the perspective of prediction accuracy, the energy industry and healthcare industry have relatively high rates, reaching 97.16 and 95.77%, respectively, while in the transportation industry, they have reached 92.87%, and in agriculture, they have reached 88.34%. From the perspective of prediction efficiency, it is relatively high in the transportation and healthcare industries, both reaching over 80%, while it is also above 75% in the energy and agriculture industries. From the perspective of resource utilization rates in the four industries, the energy industry, transportation, agriculture, and healthcare have all reached over 85%. In summary, the combination of artificial intelligence and new real-time disassembly sensors has been extended to the energy industry, transportation industry, agriculture, and medical and health industries for collaborative effect verification after predicting generator failures in the manufacturing industry. It has been found that the predictive performance has reached good levels, indicating that the application scope has been expanded to a certain extent.

5.1 Comparison of performance parameters for generator fault prediction

After the above experiment, the accuracy, precision, recall, and F1 value of the generator fault prediction performance are compared and analyzed, as shown in Figure 2.

Figure 2

Performance parameters for predicting generator faults.

From Figure 2, it can be seen that after tenfold cross-validation, the accuracy of this experiment reached an average of 0.94, and the precision reached an average of 0.92. The average recall rate reached 0.96, and the average F1 value reached 0.87. Overall, the generator fault prediction performance has achieved good predictive performance in terms of accuracy, precision, recall, and F1 value.

5.2 Detection and maintenance performance of generator faults

To further verify the synergistic effect of artificial intelligence technology and new real-time disassembly sensors, the detection and maintenance performance of generator faults is now verified as shown in Figure 3. Among them, four types can be selected for comparison, namely optimized GRU – a new real-time disassembly sensor, random forest – a new real-time disassembly sensor, new real-time disassembly sensor, and traditional sensor.

Figure 3

Detection and maintenance performance of generator faults.

In Figure 3, in terms of detection efficiency, the random forest new real-time disassembly sensor achieved 89.35%. After optimization, the GRU new real-time disassembly sensor achieved 94.38%, which is 5.03% higher than the random forest new real-time disassembly sensor. It can be seen that the optimized GRU model in this experiment has a better synergistic effect compared to the random forest model and the new real-time disassembly sensor and has better efficiency in generator fault detection. In addition, the detection efficiency of using the new real-time disassembly sensor alone is only 85.90%, which is 8.48% lower than the optimized GRU new real-time disassembly sensor. It can be seen that without the introduction of artificial intelligence technology models, it is difficult to achieve predictive maintenance, there is a certain delay, and the detection efficiency is seriously reduced. The limitations of traditional sensors are more obvious, with a detection efficiency of only 69.24%. Compared with the optimized GRU new real-time disassembly sensor, it has decreased by 25.14%, with a larger amplitude. From the perspective of cost loss rate, the optimized GRU new real-time disassembly sensor is only 3.03%, while the new real-time disassembly sensor reaches 8.94%. In terms of utilization, the optimized GRU new real-time disassembly sensor reached 86.09%, while the new real-time disassembly sensor only reached 71.25%. Due to serious reasons such as downtime delay, the resource utilization rate has decreased.

Overall, it is difficult to achieve predictive maintenance and significantly increase downtime costs for a single new type of dismantling sensor, which has significant limitations. However, the application of artificial intelligence technology, such as optimized GRU models and random forest models, has to some extent overcome the limitations of a single new type of real-time dismantling sensor.

5.3 User satisfaction

User satisfaction analysis is the process of evaluating the level of satisfaction of users with products, services, or experiences. This experiment evaluates the optimized GRU – a new real-time disassembly sensor, random forest – a new real-time disassembly sensor, a new real-time disassembly sensor, and a traditional sensor through the collection of survey questionnaires. The specific evaluation indicators are maintenance timeliness, usability, detection accuracy, and stability. These four indicators are all 10 points, a total of 40 points. A total of 318 questionnaires were collected, and the user satisfaction analysis results after mean processing are shown in Figure 4.

Figure 4

User satisfaction results.

In Figure 4, overall, the optimized GRU new real-time disassembly sensor achieved a score of 34, accounting for 85% of the total score. The total score of the random forest – a new real-time disassembly sensor reached 31 points, accounting for 77.5% of the total score. The new real-time disassembly sensor scored 25 points, accounting for only 62.5% of the total score. Compared to other sensors, traditional sensors are less satisfactory to users, accounting for only 40%. Overall, in the era of artificial intelligence technology, the optimized GRU new real-time disassembly sensor is more favored by users, while individual sensors are gradually withdrawing from the trend of the times.