Konsol Onsite EEWS: pemicu Sa(T) dari gelombang P awal
Konsol ini memutar ulang rekaman gempa nyata dari arsip lewat 12 node onsite dan menampilkan keputusan peringatan dini dari satu stasiun: spektrum Sa(T) terprediksi, level aksi, rasio terhadap spektrum desain, dan waktu yang tersisa sebelum puncak guncangan. Petunjuk ini menjelaskan cara memakainya.
🎯
1. Tentang konsol ini
Satu stasiun akselerograf menilai guncangan dari beberapa detik pertama gelombang P, sebelum gelombang S yang merusak tiba. Konsol ini memperagakan bagaimana keputusan seperti itu dibuat dan dibaca.
Masukan: rekaman akselerasi 3 komponen dari arsip, diumpankan ke node paket demi paket (1 detik).
Proses di node: deteksi gelombang P, pengambilan jendela 3 detik, prediksi Sa(T) pada 102 periode.
Keluaran: level aksi, rasio D/C terhadap spektrum desain SNI 1726, dan lead time ke puncak percepatan.
Tekan salah satu tombol replay (S1, S3 atau S5) di kanan bawah peta.
Buka kejadian yang muncul pada log di sebelah kanan.
Baca grafik spektrum (prediksi terhadap desain), akselerogram 3 s, lead time dan empat kartu ringkas.
Sentuh ikon ⓘ mana saja untuk penjelasan istilah dalam popup.
Skenario
Apa yang terlihat
S1
Gempa Cianjur 2022 (M 5,6), stasiun DSJR, 11 km. Level A; lead time ke puncak 6,38 s; Sa(0,2 s) prediksi 59,9 gal terhadap 243 gal teramati. Rekaman ini ikut melatih model, jadi hanya ilustrasi alur.
S3
Rekaman uji (M 7,2, 22 Juli 2026), stasiun BMASI, 104 km; model tidak pernah melihatnya. Level A; lead time 17,65 s; Sa(0,2 s) prediksi 32,3 gal terhadap 35,2 gal teramati.
S5
Prediksi kurang (M 5,3, 94 km, stasiun BSJR): picker tepat waktu, tetapi Sa(0,2 s) prediksi 47,5 gal terhadap 84,0 gal teramati; keduanya di bawah ambang Level A (131,3 gal), level tetap STANDBY.
Satu replay hanyalah contoh. Ia tidak mengukur kinerja sistem secara keseluruhan.
🗺️
3. Anatomi antarmuka
Apa yang ditampilkan dan apa artinya
Elemen
Keterangan
HUD status
Bar di atas peta: menunjukkan status siaga atau keputusan terakhir (level aksi, D/C) dan lead time hasil replay, yaitu selisih keputusan dan puncak percepatan horizontal (bukan hitung mundur dan bukan lead time ke gelombang S).
Peta 12 stasiun
Setiap penanda adalah satu node onsite. Klik untuk melihat stasiun, objek yang dilindungi, kelas situs, parameter SNI (SDS, SD1, Ta) dan status node.
Tombol skenario
Tiga replay rekaman arsip nyata melalui node dan model yang sama: S1, S3 dan S5.
Modal analisis
Spektrum Sa(T) prediksi terhadap spektrum desain SNI 1726 situs, akselerogram 3 komponen jendela 3 s, level aksi, D/C, lead time, dan aliran waktu. Geser dasar, drift, dan kerusakan ditampilkan n/a karena tidak dihitung.
Papan relai SCADA
Peragaan tampilan saja; aplikasi tidak terhubung ke relai atau aktuator.
Header
Tombol tema gelap/terang, bahasa ID/EN, dan tombol cara membaca (ikon ⓘ).
D/C. Sa(Ta) prediksi dibagi Sa desain pada Ta (penjelasan D/C ⓘ). Ini rasio terhadap spektrum desain, bukan penilaian aman atau tidak aman.
Lead time. Selisih antara keputusan (t_P + 3 s) dan puncak percepatan horizontal (penjelasan lead time ⓘ); aliran waktu ⓘ menunjukkan komponennya.
Level aksi. Pemetaan ilustratif dari keluaran model ke contoh tindakan (penjelasan level aksi ⓘ). Level tertinggi yang terpenuhi menjadi keputusan; bila tidak ada, statusnya STANDBY.
Level
Contoh tindakan
Ambang
A
Gas / utilitas
Sa(0,2 s) ≥ 131 gal, atau P(III+) ≥ 0.446, atau D/C ≥ 0,50
B
Lift ke lantai terdekat
Sa(0,5 s) ≥ 75 gal, atau D/C ≥ 0,75
C
Kunci peredam MR
Sa(1,0 s) ≥ 22 gal, atau D/C ≥ 1,00
D
Rem kereta cepat
P(IV+) ≥ 0.351, atau D/C ≥ 1,25
⚠️
5. Batasan yang perlu diketahui
Replay, bukan langsung. Node belum tersambung ke aliran SeedLink langsung; semua keputusan yang tampil berasal dari rekaman arsip.
Satu situs. Setiap keputusan berasal dari satu stasiun; konsensus antar stasiun tidak diimplementasikan.
Level aksi ilustratif. Ambang A–D ditetapkan pada data latih dan tidak diuji sebagai aturan operasional; tidak ada aktuator terhubung.
Jendela tetap 3 s. Periode panjang (T ≥ 1 s) kurang didukung oleh jendela sependek itu, dan prediksi cenderung terlalu rendah pada gempa besar.
Tanpa respons struktur. Geser dasar, drift, kerusakan dan evakuasi tidak dihitung. Nama fasilitas, tinggi dan Ta adalah asumsi ilustratif.
Gambar fasilitas. Hanya 3 fasilitas memakai foto nyata berlisensi terbuka; sisanya skema, ditandai demikian di popup.
Waktu dari keputusan ke puncak percepatan horizontal; bukan ke gelombang S.
D/C
Sa(Ta) prediksi dibagi Sa desain SNI 1726 pada periode Ta.
Ta
Periode getar fundamental bangunan ilustratif di stasiun.
Level aksi
Contoh tindakan A–D yang dipicu oleh keluaran model; ilustratif.
STANDBY
Keputusan tanpa level aksi.
GPD
Generalized Phase Detection (Ross dkk. 2018): jaringan saraf yang memberi peluang P pada jendela 4 s; dijalankan kausal di node.
Rekaman latih / uji
Latih: sebelum 1 April 2026, dipakai membangun model. Uji: sesudahnya, tidak pernah dilihat model.
Replay
Rekaman arsip nyata yang diumpankan ke node; bukan gempa yang sedang terjadi.
📖 USER GUIDE
Onsite EEWS console: Sa(T) trigger from the early P wave
This console replays real archived earthquake records through 12 onsite nodes and shows the early-warning decision from a single station: the predicted Sa(T) spectrum, the action level, the ratio to the design spectrum, and the time left before peak shaking. This guide explains how to use it.
🎯
1. About this console
A single strong-motion station assesses ground shaking from the first seconds of the P wave, before the damaging S waves arrive. This console demonstrates how such a decision is made and how to read it.
Input: 3-component acceleration records from the archive, streamed to the node in 1-second packets.
Processing at the node: P-wave detection, a 3 s window, Sa(T) prediction at 102 periods.
Output: action level, the D/C ratio to the SNI 1726 design spectrum, and the lead time to peak acceleration.
Press one of the replay buttons (S1, S3 or S5) at the lower right of the map.
Open the event that appears in the log on the right.
Read the spectrum chart (predicted against design), the 3 s accelerogram, the lead time and the four summary cards.
Tap any ⓘ icon for an explanation of the term in a popup.
Scenario
What you see
S1
Cianjur 2022 earthquake (M 5.6), station DSJR, 11 km. Level A; lead time to the peak 6.38 s; predicted Sa(0.2 s) 59.9 gal against 243 gal observed. This record was part of the training data, so it only illustrates the flow.
S3
Test record (M 7.2, 22 July 2026), station BMASI, 104 km; the model never saw it. Level A; lead time 17.65 s; predicted Sa(0.2 s) 32.3 gal against 35.2 gal observed.
S5
Under-prediction (M 5.3, 94 km, station BSJR): the picker is on time, but predicted Sa(0.2 s) is 47.5 gal against 84.0 gal observed; both are below the Level A threshold (131.3 gal), so the level stays STANDBY.
A single replay is only an example. It does not measure the overall performance of the system.
🗺️
3. Interface anatomy
What is shown and what it means
Element
Description
Status HUD
Bar above the map: shows the standby state or the latest decision (action level, D/C) and the replay's lead time, i.e. the gap between the decision and the peak horizontal acceleration (not a countdown, and not a lead time to the S-wave arrival).
12-station map
Each marker is one onsite node. Click for the station, the protected facility, site class, SNI parameters (SDS, SD1, Ta) and node status.
Scenario buttons
Three replays of real archived records through the same node and model: S1, S3 and S5.
Analysis modal
Predicted Sa(T) against the site SNI 1726 design spectrum, 3-component accelerogram of the 3 s window, action level, D/C, lead time and time flow. Base shear, drift and damage show n/a because they are not computed.
SCADA relay board
Display demonstration only; the app is not connected to relays or actuators.
Header
Buttons for dark/light theme, ID/EN language and the how-to-read guide (ⓘ icon).
📈
4. Reading the results
Click the links marked ⓘ for an explanation in a popup
D/C. Predicted Sa(Ta) divided by the design Sa at Ta (explain D/C ⓘ). It is a ratio to the design spectrum, not a safe or unsafe verdict.
Lead time. The difference between the decision (t_P + 3 s) and the peak horizontal acceleration (explain lead time ⓘ); the time flow ⓘ shows its components.
Action level. An illustrative mapping from model output to example actions (explain action levels ⓘ). The highest level met becomes the decision; if none is met the status is STANDBY.
Level
Example action
Threshold
A
Gas / utilities
Sa(0.2 s) ≥ 131 gal, or P(III+) ≥ 0.446, or D/C ≥ 0.50
B
Lift to nearest floor
Sa(0.5 s) ≥ 75 gal, or D/C ≥ 0.75
C
MR damper lock
Sa(1.0 s) ≥ 22 gal, or D/C ≥ 1.00
D
High-speed-rail brake
P(IV+) ≥ 0.351, or D/C ≥ 1.25
⚠️
5. Limits to keep in mind
Replay, not live. The nodes are not yet connected to the live SeedLink stream; every decision shown comes from archived records.
Single site. Each decision comes from one station; multi-station consensus is not implemented.
Illustrative action levels. The A–D thresholds were set on training data and are not tested as an operational rule; no actuator is connected.
Fixed 3 s window. Long periods (T ≥ 1 s) are poorly supported by so short a window, and the prediction tends to be too low for large earthquakes.
No structural response. Base shear, drift, damage and evacuation are not computed. Facility names, heights and Ta are illustrative assumptions.
Facility images. Only 3 facilities use real openly licensed photographs; the rest are schematics, labelled as such in the popup.
Pseudo-spectral acceleration (PSA) of a 5%-damped oscillator with natural period T.
θ
Trigger threshold on Sa(T), in gal.
Lead time
Time from the decision to the peak horizontal acceleration; not to the S-wave arrival.
D/C
Predicted Sa(Ta) divided by SNI 1726 design Sa at period Ta.
Ta
Fundamental natural period of the illustrative building at the station.
Level aksi
Example actions A–D triggered by the model output; illustrative.
STANDBY
A decision with no action level.
GPD
Generalized Phase Detection (Ross et al. 2018): a convolutional neural network that classifies a 4 s three-component window as P-phase, S-phase or noise; run causally in the node on streaming data.
Rekaman latih / uji
Training: before 1 April 2026, used to build the model. Test: later, never seen by the model.
Replay
A real archived record fed to the node; not an earthquake in progress.
