A city gas distribution system runs on high-, medium-, and low-pressure pipelines that crisscross the city, and the effect of a single leak often does not stop at the leak point itself. When corrosion pierces a high-pressure pipeline, the gas vents to the atmosphere while the outlet pressure of the downstream regulator station starts to fall, a low-pressure alarm triggers at the far end of the medium-pressure network, the emergency shutoff valve trips, and tens of thousands of households lose supply. From the leak to the shutdown, the whole sequence can take less than 50 minutes. The monitoring wall flashes plenty of alarms, but the questions operators actually need answered are often the same three: Where is the leak? Is the pressure drop a leak or peak-hour demand? Is the abnormal regulator outlet a device fault or upstream propagation? Answering these does not call for more alarms. It calls for data that can reconstruct the full causal propagation chain.
Pipeline cascade: the hardest part of gas distribution
The difficulty of city gas distribution comes down to a single word: linkage.
From the gate station to the customer, natural gas passes through three pressure steps: high pressure at 1.0-1.2 MPa, medium pressure at 0.20-0.30 MPa, and low pressure below 0.01 MPa. At each step down, the regulator handles the reduction and the emergency shutoff valve acts as the last line of defense. The system runs steadily; the problem is that a single failure propagates through the whole chain.
When corrosion causes a through-wall leak in a section of the high-pressure network, the pressure drop propagates downstream along the line. The gate station SCADA detects the drop and raises feed flow, but the leak outruns the feed capacity and the outlet pressure keeps falling. The regulator senses the low inlet pressure and pushes its opening wider trying to hold the outlet pressure, but the high-pressure side has nothing left to give. Once the outlet pressure falls through the interlock threshold, the emergency shutoff valve closes and the downstream medium-pressure network loses supply instantly. From the corrosion-through on the high-pressure section to the medium-pressure shutdown, 50 minutes pass, and the propagation chain spans 6 devices.
What makes it worse is the physical distance between devices. One city’s gas distribution system manages 85 km of high-pressure pipeline, 1200 km of medium-pressure pipeline, and 24 regulator stations. The leak detectors are battery-powered IoT devices that wake only once every 2 hours, so detection points are inherently sparse. When a leak happens between inspection rounds, inferring the location from the pressure gradient is nearly the only option, and pressure is distorted by the peaks and valleys of demand, with normal daily swings of ±35%, so the signal-to-noise ratio is extremely low.
This is the hardest part of gas distribution monitoring: anomalies cascade from device to device, while detection is both sparse and slow. The operations team has to pull the signal out of the noise and then string the signals into a causal chain.
Operations pain points: many alarms, few answers
Pressure dropped: is it a leak or peak-hour demand?
The same symptom, a falling pipeline-section pressure, can have completely different root causes.
During peak demand, the east gate station’s instantaneous flow jumps from 9000 to 15000 m³/h, and network pressure naturally dips by 5%-10%. That is a normal load response. But the same pressure drop, when it comes from a corrosion-through leak on the high-pressure network, sees PT-301 section pressure slide from 1.18 MPa down to 0.45 MPa, a fall of 62%. The first case needs no action; the second triggers an emergency shutdown 50 minutes later.
The problem is that the pressure drop in the early stage of a leak has a very gentle slope. At the start of dispatch cycle D-088, PT-301 pressure only falls from 1.18 to 1.05 MPa, an 11% change that sits entirely inside the normal fluctuation band. A dispatcher glancing at the wall can hardly tell whether the evening peak is arriving or a section is failing. By the time leak detector LD-301 reads a combustible gas concentration of 210 ppm, the pressure has already fallen further.
Looking at pressure alone, an early leak and peak-hour demand are nearly indistinguishable. You have to overlay three sets of data: pressure, flow, and leak concentration. While pressure falls, flow rises (SCADA auto-boosts the feed) and concentration rises too. Only when all three indicators move together can you confirm a leak rather than a load swing.
Regulator outlet overpressure: is the problem the regulator or upstream?
The west city regulator RG-501 shows an outlet pressure of 0.33 MPa, above the 0.30 MPa upper limit, so a regulation-abnormal alarm triggers. The dispatcher’s first reaction is that the regulator itself has failed. But outlet overpressure has at least two possible causes: a fault in the regulator body that opens it too far, or an abnormal rise in upstream high-side pressure that breaks the reduction ratio.
A single-point alarm cannot tell the two root causes apart. If you misread it as an upstream problem when it is actually an aging, sticking membrane, you miss the repair window. The diaphragm health index falls from 0.83 to 0.72, below the 0.80 replacement threshold, the sticking keeps worsening, the outlet pressure climbs from 0.33 MPa to 0.52 MPa, 86% above setpoint, the entire medium-pressure network runs overpressure, and joints on the customer side start to leak.
Misreading it the other way sends a repair crew to the regulator station for nothing when the real cause is upstream pressure propagation. The key is to watch two device-level indicators: the regulator opening and the diaphragm health index. If the opening rises abnormally while the membrane health drops, the regulator itself is at fault. If the opening is normal but the inlet pressure is high, it is upstream propagation.
After a leak alarm, how wide should the shutoff be?
When a high-pressure network leak alarm triggers, the emergency dispatch center’s first decision is whether to close the emergency shutoff valves, and which ones.
Closing east city ESV-401 cuts supply to 32,000 households. Leaving it open lets the leak continue, venting gas and raising the safety risk. If the west city direction is also affected, does ESV-501 need to close? That depends on where the leak is and how the network is wired.
But the alarm system only tells you that LD-301 concentration is over the limit. It does not tell you the leak location, its scale, or its direction of spread. The operator has to manually pull up the PT-301 pressure trend, the FT-101 flow compensation curve, and the PT-102 outlet pressure changes, then rely on experience to judge whether the leak is in the PT-301 section, roughly how large it is, and whether a regulator station will be affected. This whole process is data assembly by hand, and in a response window measured in minutes, one wrong step means a wide-area outage or a safety incident.
From passive monitoring to proactive judgment
Object-based modeling: writing device relationships into the asset tree
The gas distribution system is not a collection of 15 independent devices. It is a causal chain that runs gate station → high-pressure network → regulator station → medium-pressure network → customer end. Object-based modeling writes the structure of that chain into the asset tree.
A city gas distribution system’s asset tree is organized into four sections: the gate station section (PT-101/102, FT-101, and other inlet and outlet devices), the pipeline section (PT-301/601 pressure transmitters, LD-301/601 leak detectors), the regulation section (RG-401/501 regulators, ESV-401/501 emergency shutoff valves), and the emergency dispatch center (WS-701 environmental monitoring station). Each device’s position in the asset tree encodes its upstream and downstream relationship in the distribution chain. PT-301 sits in the pipeline section on the high-pressure side, and RG-401 sits in the regulation section on the east city side. The asset tree hierarchy is a map of causal propagation.
The benefit of this modeling is that when an alarm triggers, the operator does not have to look up who sits upstream and downstream of a device by hand. The asset tree already defines the topology. When PT-301 pressure falls, the system links LD-301 in the same section and RG-401 in the downstream regulation section, making the propagation path explicit.
Real-time analysis and event correlation: from single-point alarms to capturing propagation chains
A single-point alarm can only answer where the problem is. Event correlation answers where it came from and where it is going.
Take a high-pressure network leak as an example. IDMP configures three layers of alarm correlation. The first layer is standing alarms: PT-301 section pressure below 0.95 MPa for 10 minutes triggers a Critical alarm, LD-301 combustible gas concentration above 500 ppm for 5 minutes triggers a Critical alarm, and RG-401 outlet pressure below 0.18 MPa for 5 minutes triggers a Critical alarm. These three alarms are not isolated. PT-301 is the root cause, LD-301 is the direct evidence, and RG-401 is the interlock early-warning point.
The second layer is trace-back indicators that the operator expands step by step after an alarm. First confirm whether the LD-301 concentration is synchronized with the PT-301 pressure drop (lag no more than 5 minutes). Then pull up the FT-101 flow curve to size the leak (compensation flow above 4000 m³/h means a large leak). Finally, check the PT-102 outlet pressure and the RG-401 outlet pressure to predict the downstream impact. Three alarm layers plus two trace-back layers string together the full analysis chain of detection → location → assessment → prediction.
Process event capture pushes correlation further. Pipeline leak events, regulator overpressure events, and high-pressure network pressure plunge events. These event windows record not only the trigger and recovery times but also the extreme values of the key indicators during the event, giving post-incident review structured data.
Process analysis and AI-assisted insights: spotting the signal in the noise early
Gas distribution has a strong daily cycle. Flow peaks twice, from 6 to 9 in the morning and 5 to 9 in the evening, and the off-peak flow is only half the peak flow. This regular noise buries the abnormal signal. In the early stage of a corrosion-through on the high-pressure network, PT-301 pressure drifts down slowly from 1.18 to 1.05 MPa, an 11% change that nearly overlaps the normal peak-to-trough swing.
Process analysis separates the trend change from the cyclical swing. In a normal daily cycle, PT-301 section pressure sits in hold mode, a horizontal line around 1.18 MPa with noise no larger than ±0.012 MPa. When a leak starts, the pressure trend turns from a horizontal line into a downward slope. Even with a gentle gradient, the directional shift is statistically significant. AI-assisted insights can detect this trend shift on a sliding window and raise an alert within 5 minutes of the leak, far earlier than the concentration alarm.
The longer-term value is device health trend analysis. The west city regulator RG-501’s diaphragm health index falls from 0.95 to 0.72 over 7 days. Since the D-062 maintenance was deferred, the membrane health has been slowly declining. If the system could raise a Warning when the diaphragm health index falls below 0.80 and prompt the operations team to schedule a membrane replacement, the stuck-valve overpressure incident would never happen. This is the critical step from post-incident review to advance prediction.
The analysis closed loop in typical abnormal scenarios
Case 1: cascading failure from a corrosion-through leak on the high-pressure network
At 4 PM on Day -4, the section holding PT-301 on the high-pressure network (DN600, 14 years in service) has corroded internally down to 3.2 mm. Combined with a persistently low protective potential on that section of the cathodic protection system, the corrosion accelerates and pierces the pipe. Gas leaks through the hole, triggering a cascading failure that spans 8 dispatch cycles and ends with the east city medium-pressure network losing supply for 42 minutes, affecting 32,000 households.
Step 1: leak detection, a pressure downslope in resonance with a rising concentration
On the IDMP pressure monitoring panel, PT-301 section pressure enters a sustained downslope starting at 16:03 on D-088, falling from 1.18 MPa to 1.05 MPa. Compared with the horizontal hold-mode line of a normal daily cycle, the directional shift is clear. Five minutes later, LD-301 combustible gas concentration rises from 0 to 210 ppm, matching the timing of the PT-301 pressure drop. Pressure falling and concentration rising at the same time places the leak tentatively in the PT-301 section.
Step 2: pressure confirmation, flow compensation verifies the leak signature
At 17:12 on D-089, FT-101 instantaneous flow rises from 12500 to 16500 m³/h, up 32%. The gate station SCADA detects the high-side pressure drop and raises the feed, yet the PT-102 outlet pressure keeps falling, from 1.16 to 0.88 MPa. Feed rising upstream while pressure still falls downstream is the classic leak signature. Peak-hour demand is ruled out, and the leak is sized as large, with compensation flow above 4000 m³/h.
Step 3: downstream propagation, outlet pressure loss approaches the interlock threshold
At 18:22 on D-090, the east city regulator RG-401’s inlet pressure falls from 1.16 to 0.62 MPa. The regulator opens fully, from 42% to 95% opening, but still cannot keep up, and the outlet pressure falls from 0.282 to 0.21 MPa. By 19:05 on D-091, the outlet pressure is down to 0.16 MPa, close to the 0.15 MPa interlock threshold.
Step 4: emergency interlock, ESD triggers and root cause trace-back
At 19:08 on D-091, ESV-401 switches from remote control to automatic interlock. At 19:09 the emergency shutdown state is set to true, the valve opening drops from 100% to zero, and the east city medium-pressure network loses supply. In the post-incident trace-back, the historical cathodic protection data confirms that the protective potential on this section sat below the -0.85V standard for a long time, and accelerated corrosion was the root cause of the puncture. From the first PT-301 pressure drop to the ESD trigger, 50 minutes elapsed. If the alarm correlation and trace-back paths had been configured in advance, at least 30 minutes of early-warning time could have been gained.
Case 2: overpressure from an aged, sticking regulator diaphragm
At 8 AM on Day -1, the main membrane of the west city regulator RG-501 (8 years in service) ages and the valve stem sticks in the high position. Worse, this regulator’s scheduled weekly maintenance on Day -5 (D-062~D-064) was postponed because crews were busy on other tickets, so the membrane replacement never happened. The sticking hazard builds for 7 days before it erupts. The outlet pressure runs continuously over the limit, the medium-pressure network goes overpressure, joints on the customer side start to leak, the emergency shutoff valve trips, and the west city medium-pressure network loses supply for 38 minutes, affecting 28,000 households.
Step 1: outlet overpressure, an abnormal opening in resonance with falling membrane health
At 08:05 on D-152, RG-501’s opening rises from 42% to 58%, up 38%. At 08:10 the outlet pressure rises from 0.282 to 0.33 MPa, above the 0.30 MPa upper limit. A regulation-abnormal alarm triggers, and the dispatcher writes it off as a swing. But when the opening and diaphragm health index trends are overlaid on the IDMP panel, the picture is different. The opening rises abnormally while the diaphragm health index falls from 0.95 to 0.83. The two moving together point to a fault in the regulator body, not upstream propagation.
Step 2: membrane health, tracing 7 days of gradual aging
At 09:08 on D-153, the diaphragm health index falls to 0.72, below the 0.80 replacement threshold. Tracing the diaphragm health index history shows that since the D-062 maintenance was deferred, membrane health has declined steadily, 0.95 → 0.83 → 0.72, accumulating over 7 days. If the system had raised a Warning when the diaphragm health index fell below 0.80 and reminded the team to schedule a replacement, this overpressure incident could have been avoided entirely.
Step 3: ESD interlock, overpressure propagates to emergency shutdown
At 10:08 on D-154, PT-601 pressure on the medium-pressure network rises to 0.43 MPa. At 10:14, LD-601 detects a combustible gas concentration of 680 ppm, as the customer-side joints before the meter start to leak under the overpressure. At 11:10 on D-155, the medium-pressure network pressure exceeds the 0.45 MPa interlock threshold, ESV-501’s emergency shutdown state is set to true, and the west city medium-pressure network loses supply. From the first abnormal RG-501 opening to the ESD trigger, 55 minutes pass. If the diaphragm health index warning had fired in the D-062 maintenance window, the warning lead time could have been stretched to 7 days.
Closing Thoughts
For safety monitoring in a city gas distribution system, the core problem is not a shortage of data. It is that the causal chains between data points are unclear. Pressure drops, flow rises, concentration over-limits, ESD trips. Operators see these alarms every day, but working out the propagation paths and timing between them by hand on a monitoring wall is slow and error-prone.
IDMP writes the topology of that causal chain, gate station → high-pressure network → regulator station → medium-pressure network → customer end, into the asset tree. It configures the three alarm correlation layers and the two trace-back paths as a standard workflow, and it lifts device health trends out of the noise to become early-warning signals. Operators no longer assemble the puzzle by hand inside an alarm storm. They follow the propagation path the system lays out, tracing from root cause to the end state, and move from post-incident review to advance prediction.
TDengine comes with a high-performance, distributed time-series database, Industrial Ontology modeling, and an Industrial Agent Runtime, providing a full-stack solution for industrial data streams from collection and storage to real-time analytics, visualization, event management, and root-cause analysis. To learn more about TDengine, visit www.tdengine.com and try it for free.
Try it yourself
Install and deploy TDengine Visit the TDengine Download Center, select TDengine All-in-One, choose the deployment platform and architecture that matches your environment, and follow the guided steps to complete the installation.
Load the sample data
On first activation, choose Urban Gas Safety Monitoring on the sample data loading screen and wait for it to finish loading.
If you have already activated the product, click your avatar in the top-right corner, select Management Console, choose Sample Data on the left, then select Urban Gas Safety Monitoring. Wait a few minutes for the data to load.
Urban Gas Safety Monitoring.
