In the non-ferrous metal smelting industry, high-temperature molten
pool composition fluctuations, electrolyzer plate corrosion, casting
defects detection and other process challenges, long-term reliance
on manual experience judgment, resulting in unstable quality control,
high energy consumption. With the breakthrough of industrial camera,
multi-spectral imaging and AI algorithm, intelligent visual recognition
technology is reconstructing the whole process of non-ferrous smelting
with the “eye of the machine” - from the real-time composition analysis
in the melting furnace to the micron-level surface defects capture, from
the prediction of the health of the equipment to the process of
self-optimization of parameters. From real-time composition analysis
in the melting furnace to micron-level surface defect capture, from
equipment health prediction to process parameter self-optimization,
an industrial change driven by visual intelligence has already started.
A technological breakthrough: Intelligent vision
cracks the “invisible” problem of
non-ferrous smelting.
The process environment of non-ferrous smelting is characterized by high
temperature, high corrosivity and strong interference, which makes
it difficult for the traditional vision system to operate stably.
The new generation of intelligent vision technology realizes the
subversion through three major innovations:
Ultra-high temperature imaging technology:
Adopting short-wave infrared (SWIR) camera and active cooling
system, it can penetrate smoke and flame inside the smelting
furnace at 1200℃, and capture the flow state of copper liquid
and impurity distribution in real time, with an imaging
accuracy of 0.1mm²/pixel.
Case: A copper smelter deployed a multispectral camera
at the converter mouth, combined with a CNN algorithm
to identify Fe3O4 crystallization on the melt surface,
reducing metal loss due to peroxidation by 3.7%.
Multimodal data fusion:
Synchronized acquisition of visible light, thermal imaging,
and X-ray images, cross-modal feature extraction through
the Transformer model to accurately identify composite
defects such as electrolytic anode mud thickness and zinc
ingot subcutaneous porosity.
Edge Intelligent Computing:
The edge computing box equipped with industrial-grade
GPU can complete 1280×1024 resolution image processing
within 0.5 seconds to meet the real-time inspection demand
of 30 aluminum plates per minute in the casting line.
Second, the scene reconstruction: intelligent vision
driven by the four core process changes
1. Smelting: from “experience fire control” to “visual alchemy”.
Through the panoramic monitoring system in the furnace, real-time
analysis of copper matte molten pool color changes and bubble
morphology, AI model dynamically recommended blowing air
volume, fuel ratio and other parameters, so that the range of
fluctuations in crude copper grade from ± 2% narrowed to ± 0.5%.
The infrared thermal camera scans the inner wall of refractory
material and combines with ResNet to predict the erosion
location, reducing the maintenance response time from 72
hours to 4 hours and extending the furnace life by 20%.
2. Electrolysis: the “AI monitor” for the health of the electrodes
High-definition line array camera scans the surface of cathode copper
plate to detect 13 types of defects such as dendritic crystallization and
pockmarks, and the detection rate has increased from 82% of manual
visual inspection to 99.6%, avoiding tens of millions of quality claims
every year.
Multi-spectral imaging system monitors the corrosion morphology of
anode plate and predicts the remaining life through LSTM algorithm,
and the annual power saving of a single tank after optimizing the
replacement cycle of the plate reaches 42,000 kWh.
3. Pouring process: zero-tolerance defense against micron-level defects
3D structured light camera scans the surface of zinc ingot, reconstructs the
3D model with 0.01mm precision, identifies cracks, slag and other defects,
and reduces the defect rate from 1.2% to 0.15%.
High-speed vision system with robotic arm, real-time correction of
aluminum bar casting offset, dimensional tolerance control within ±
0.3mm, directly through the high-end aviation aluminum
certification threshold.
4. Environmental protection control: “full-time eye in the
sky” of the emission source.
The UV imager is deployed at the flue gas discharge port to dynamically
monitor the concentration of SO2 and particulate matter, and the data
real-time linkage DCS system adjusts the dosage of desulfurizer, which
reduces the incidents of emission exceeding the standard by 90%.
Intelligent inspection drones equipped with hyperspectral cameras in
the plant identify hidden pipeline leakage points, reducing the loss
of metal solution leakage by more than 8 million yuan per year.
Value fission: from “quality improvement and cost reduction”
to “process reengineering”.
The dimension of quality control has been upgraded:
After a lead-zinc smelting enterprise introduced a vision system,
the surface qualification rate of cathode zinc flakes jumped from
93% to 99.8%, reducing quality losses by 120 million yuan per year.
Double reduction of energy consumption and material consumption:
Intelligent dosing system based on visual feedback
of melting pool reduces the amount of copper
concentrate reducing agent by 18%, and the
comprehensive energy consumption of tons of copper drops by 14%.
Process knowledge digitization:
Accumulating millions of smelting process image
data sets, constructing process parameter
optimization models, and realizing the
standardized inheritance of the “master's experience”.
Zero safety risk:
Replacement of manual close observation of high-temperature
melt, toxic gas environment operations,
to avoid more than 20 major safety accidents.
Fourth, the implementation path: non-ferrous enterprises
landing visual intelligence triple leap
Hardware selection and weather resistance transformation:
For the melting area (> 800 ℃), electrolysis area (strong acid fog)
and other scenes, choose explosion-proof, corrosion-resistant
camera module, with air-cooled / water-cooled protective cover.
Data closed-loop construction:
Open up the data interface between vision system and PLC, MES,
and establish the millisecond response chain of “image
acquisition - defect classification - process adjustment”.
Upgrade of human-machine cooperative system:
Develop visualized decision boards, convert AI inspection results into operation
guidelines, and cultivate new industrial workers who “know how to read data and use AI”.
V. Future picture: the deep integration of visual
intelligence and industrial meta-universe
Holographic smelting:
Through AR glasses superimposed on the virtual image,
the operator can observe the reaction process inside
the electrolysis tank through the lens, realizing “what you see is what you get” process control.
Autonomous optimization system:
Visual data stream is injected into the digital twin in real
time, and AI autonomously simulates hundreds of
parameter combinations to continuously approach the process limit.
Visual interconnection of industrial chain:
From mine ore identification to end product traceability,
visual intelligence will run through the entire non-ferrous
industry chain, building a quality and credible industrial
blockchain network.
Conclusion
Intelligent visual identification is pushing non-ferrous metal
smelting from the era of “black-box operation” to a new era of
“transparent manufacturing”. This technology not only solves
the industry's hundred years of pain, but also opens up the
infinite possibilities of process innovation - when each cluster
of flames jumping, each piece of metal crystallization by AI
accurate analysis, the non-ferrous metal industry “smart change”
tipping point has arrived. In the future, with the cross-fertilization
of machine vision and quantum sensing, brain-computer interface
and other technologies, an intelligent smelting ecology of full
perception, self-decision-making and zero defects is accelerating.