By practicing the technological system with a mainline consisting of big data analysis, intelligent diagnosis of smelting, artificial intelligence algorithm and expert control strategy, through the deep integration of Information Technology and Operational Technology, a closed-loop intelligent control system for heat state of blast furnace was developed and successfully used to serve Taigang 4 350 m³ blast furnace. The closed-loop distribution logic of such a system was designed; guided by the system development goal and technological framework, a technological route was formed, which integrated modules including material tracking, goal setting, fuel ratio optimization, data preprocessing, and abnormal situation processing. After putting into service, the production foreman only need to set target range of Si and upper/lower limits of targeted molten iron temperature, then the system can automatically track the coke load changes of blast furnace, automatically optimize targeted fuel ratio, automatically balance the feeding speed, automatically balance the key factors affecting heat state of blast furnace (such as the blast furnace heat load, the gas utilization, the quality change of coke and coal, the change of wax and wane in the hearth), and automatically calculate the coal injection quantity under the current status. These can automatically be executed into the primary system and used to replace manual operation. The application effects show that such a system can not only effectively reduce the fluctuation of furnace temperature, improve the quality of molten iron and reduce the times of double slag smelting by 30%, but also effectively promote the smooth running of blast furnace condition by stabilizing the furnace temperature and significantly reduce the fuel ratio. The comprehensive efficiency is 12.76 million yuan per year.

Hardware industry is a traditional labor-intensive industry, and its assembly line needs a lot of human resources. The problems such as high labor cost and low production efficiency seriously restrict the development of the hardware industry. A set of automatic assembly production line system for hardware industry based on machine vision was developed. Taking the 17PC toolkit assembly line as an example, the main working procedures of the assembly line were analyzed emphatically, and the design scheme of single-station modular robot workstation was put forward aiming at the automatic assembly of working procedures of components such as wrenches, connecting rods and head-marking. Combining with the characteristics of one-tow-four camera system, the machine vision algorithm was designed, and the image data of different stations were processed in real-time to form the logic of machine vision algorithm. The pre-production test of the assembly line shows that all performance indexes of the assembly line system meet the expected requirements, and the average production takt is 2.49 seconds/piece (28.9% shorter than the manual production takt of 3.5 seconds/piece), which can effectively help hardware enterprises to achieve cost reduction and efficiency increase.

Risk assessment is an effective strategy to identify the risks of assets in the intelligent factories and ensure the safe operation of assets in the digital manufacturing enterprises. Focusing on the factors including asset value, threat and vulnerability in the asset system structure of intelligent factory, the assessment strategy of asset value and asset security risk value was put forward. The Ordered Weighted Geometric Average (OWGA) operator was used to effectively gather data information and sort the importance of assets, so that the security risks could be sorted by calculating the weights of assets, risk probability and threat consequences, and further to improve the objectivity and measurability of risk assessment of assets. By considering the information transfer between assets and the analysis of security characteristics, the risk assessment process was formed. The Petri net topological structure of asset identification, threat identification and security policy identification was established by adopting the 4 kinds of relationship forms of mutual influence of assets called sequential, common, concurrent and secondary. An example shows that such a model can guide the complex systems such as intelligent factories to design corresponding security protection measures by using the parametric characteristics of asset importance, risk probability, threat consequences, etc., so as to ensure a high management, operation and maintenance level of assets.

Increasing the charging ratio of pellet mines has brought economic advantages and environmental benefits such as increasing the yields as well as promoting the energy conservation and emission reduction. The existing empirical and manual operation mode of pellet production can no longer meet the actual requirements of pellet production. It is necessary to make rational and quantitative research and analysis with the benefits of the development of artificial intelligence technology and intelligent equipment to finally realize the low cost, high quality, high efficiency and stability of pellet production. Starting from the quality requirements of pelletizing size, pelletizing strength and thermal burst temperature, by analyzing the pelletizing mechanism of pelletizing process and its influencing factors such as raw materials, equipment and control, the technological criteria such as the requirement of water content of the raw materials, the relationship between the granularity and the specific surface area, and the relationship between the rotating speed and the inclination angle of disc pelletizer were defined. By using the online granularity identification technology based on the image edge pickup, the granularity distribution of pellets was captured online in real time, and combined with professional judgment, an intelligent pelletizing control model was determined, taking the water addition as the main factor and the rotating speed of the disc as the auxiliary factor. The intelligent pelletizing system was put into field application. By optimizing the control logic in real time, the timeliness and accuracy of intelligent operation were improved to ensure the smooth production. The application practice shows that the qualified rate of pellets can be increased to 96.3% under the intelligent pelletizing mode, and the falling intensity of pellets can be kept at 7~10 times/pellet, which is significantly better than the manual mode. The application practice of such an intelligent pelletizing technology has achieved good results, providing stable pelletizing conditions for roasting process and laying a solid theoretical and practical foundation for the intelligent production.

Traditional vortex flowmeter has weak adaptability to pipeline environment, small measuring range ratio, and is easily interfered by various noises, so it cannot be applied to harsh industrial environments such as petrochemical and metallurgical environments. An intelligent vortex flowmeter with high accuracy, wide measuring range and strong environmental adaptability was proposed. In hardware design, MSP430 series single chip microcomputer with built-in low-power hard core fast Fourier transform (FFT) multiplier was used as the main control core, and the adaptive band-pass filtering and automatic gain control amplification were carried out on the weak signal output by piezoelectric vortex sensor to obtain a more stable analog frequency signal; the analog frequency signal was digitally collected and calculated to obtain a stable pulse signal related to the actual flow, so as to better overcome the interference of field mechanical vibration and electromagnetic signals, improve the environmental adaptability of the vortex flowmeter, and enhance the comprehensive measurement accuracy of the system. In software design, FFT algorithm was used as an auxiliary method to avoid high frequency signal section and greatly enhance the accuracy of measurement in the case of low flow rate and weak signal. The field test results show that the designed vortex flowmeter runs stably, and the relative indication error and repeatability error meet the standards of the first level meter in the national standard Vortex Flowmeter (JJG 1029—2007), so it has a very wide range of applicability in the measurement of water and gas medium vortex flow under various industrial environments.

Affected by the particularity of non-load-bearing body, there are many challenges in the mixed-line production of load-bearing body and non-load-bearing body. A processing technological scheme of using automatic guided vehicle (AGV) to realize the mixed-line assembly of chassis of load-bearing body and non-load-bearing body was put forward. Taking a project as an example, the layout of the assembly loop was designed, and the dynamic, split and integral mixed-line assembly processes were adopted to cover the assembly process flow of the load-bearing body and the non-load-bearing body in the same assembly loop. A three-drive four-follow omni-directional wheel structure layout suitable for the assembly process flow was designed, in which the requirements of heavy load assembly were ensured with the help of three-lift design. An innovative locking mechanism of AGV lifting-sliding platform was designed, so that the sliding platform could be locked in the whole lifting stroke, and both X and Y directions could be locked at the same time. The technological difficulties of transferring the AGV loaded vehicles between chassis assembly lines were solved, and the planned production takt of 75 JPH was achieved.

Traditional time series database (TSDB) has some limitations, such as low data throughput, low data compression rate, poor applicability, etc. It becomes a major challenge to improve the writing efficiency of massive data, reduce the data compression cost, and effectively support the statistical analysis of time series data. The key technologies of using time series database management system to deal with the above problems were discussed. Combined with the application and attention trend of the existing time series database management system in some industrial fields, the future development directions of time series database were prospected: strengthening the ability of aggregation computing, improving the flexibility of distributed architecture, researching and developing the technology of parameter self-tuning, transplanting to new storage devices and processors, etc.

Aiming at the problems of limited component models and poor scalability existed in the existing analog-digital hybrid circuit simulation software, an analog-digital hybrid circuit simulation interactive system with Ngspice simulation engine as its core was proposed. The interface function and encapsulation operation were used to realize data processing and calling functions, and the interface of circuit schematic diagram editing and parameter configuration was used as the interaction between users and the simulator. Users can select various components from the model component library to draw the simulation circuit diagram. The system can generate the netlist file corresponding to the circuit through circuit conversion. Ngspice completes the circuit simulation according to the user’s configurations, and graphically displays the simulation results. The synchronization algorithm was used to realize the synchronization of analog-digital mixed signals, and the integration design of analog-digital mixed circuit simulation technology was successfully solved. The test results show that the system has good performance and applicability, and the overall relative error of simulation results is only 1.3%.

Some industrial production processes have the characteristics of long production process, large amount of equipment, complicated equipment mechanism, etc., and the special process operation environment further aggravates the equipment failure rate and hazard. Taking the multi-sensor roughing mill in steel rolling process as the research object, a supervised learning algorithm based on multi-scale convolutional neural network (CNN) model was designed, and unsupervised learning algorithms such as LOF, K-means, GMM, SO-GAAL and MO-GAAL were studied and compared. Findings: 1) When the multi-scale CNN model is applied to the raw data of roughing mill, the accuracy of test set is as high as 99.91%, but the value of loss function is large; 2) Unsupervised learning algorithms evaluate the performance of equipment fault diagnosis by binary classification, among which the binary classification indexes of MO-GAAL algorithm are all above 98%, and the AUC value is as high as 0.99, which is the best among the 5 algorithms. Combining the advantages of the two algorithms, the algorithm fusion strategy is constructed and analyzed, which provides further solutions for different scenarios of roughing mill fault diagnosis.